1 ==============================
2 LLVM Language Reference Manual
3 ==============================
12 This document is a reference manual for the LLVM assembly language. LLVM
13 is a Static Single Assignment (SSA) based representation that provides
14 type safety, low-level operations, flexibility, and the capability of
15 representing 'all' high-level languages cleanly. It is the common code
16 representation used throughout all phases of the LLVM compilation
22 The LLVM code representation is designed to be used in three different
23 forms: as an in-memory compiler IR, as an on-disk bitcode representation
24 (suitable for fast loading by a Just-In-Time compiler), and as a human
25 readable assembly language representation. This allows LLVM to provide a
26 powerful intermediate representation for efficient compiler
27 transformations and analysis, while providing a natural means to debug
28 and visualize the transformations. The three different forms of LLVM are
29 all equivalent. This document describes the human readable
30 representation and notation.
32 The LLVM representation aims to be light-weight and low-level while
33 being expressive, typed, and extensible at the same time. It aims to be
34 a "universal IR" of sorts, by being at a low enough level that
35 high-level ideas may be cleanly mapped to it (similar to how
36 microprocessors are "universal IR's", allowing many source languages to
37 be mapped to them). By providing type information, LLVM can be used as
38 the target of optimizations: for example, through pointer analysis, it
39 can be proven that a C automatic variable is never accessed outside of
40 the current function, allowing it to be promoted to a simple SSA value
41 instead of a memory location.
48 It is important to note that this document describes 'well formed' LLVM
49 assembly language. There is a difference between what the parser accepts
50 and what is considered 'well formed'. For example, the following
51 instruction is syntactically okay, but not well formed:
57 because the definition of ``%x`` does not dominate all of its uses. The
58 LLVM infrastructure provides a verification pass that may be used to
59 verify that an LLVM module is well formed. This pass is automatically
60 run by the parser after parsing input assembly and by the optimizer
61 before it outputs bitcode. The violations pointed out by the verifier
62 pass indicate bugs in transformation passes or input to the parser.
69 LLVM identifiers come in two basic types: global and local. Global
70 identifiers (functions, global variables) begin with the ``'@'``
71 character. Local identifiers (register names, types) begin with the
72 ``'%'`` character. Additionally, there are three different formats for
73 identifiers, for different purposes:
75 #. Named values are represented as a string of characters with their
76 prefix. For example, ``%foo``, ``@DivisionByZero``,
77 ``%a.really.long.identifier``. The actual regular expression used is
78 '``[%@][-a-zA-Z$._][-a-zA-Z$._0-9]*``'. Identifiers that require other
79 characters in their names can be surrounded with quotes. Special
80 characters may be escaped using ``"\xx"`` where ``xx`` is the ASCII
81 code for the character in hexadecimal. In this way, any character can
82 be used in a name value, even quotes themselves. The ``"\01"`` prefix
83 can be used on global variables to suppress mangling.
84 #. Unnamed values are represented as an unsigned numeric value with
85 their prefix. For example, ``%12``, ``@2``, ``%44``.
86 #. Constants, which are described in the section Constants_ below.
88 LLVM requires that values start with a prefix for two reasons: Compilers
89 don't need to worry about name clashes with reserved words, and the set
90 of reserved words may be expanded in the future without penalty.
91 Additionally, unnamed identifiers allow a compiler to quickly come up
92 with a temporary variable without having to avoid symbol table
95 Reserved words in LLVM are very similar to reserved words in other
96 languages. There are keywords for different opcodes ('``add``',
97 '``bitcast``', '``ret``', etc...), for primitive type names ('``void``',
98 '``i32``', etc...), and others. These reserved words cannot conflict
99 with variable names, because none of them start with a prefix character
100 (``'%'`` or ``'@'``).
102 Here is an example of LLVM code to multiply the integer variable
109 %result = mul i32 %X, 8
111 After strength reduction:
115 %result = shl i32 %X, 3
121 %0 = add i32 %X, %X ; yields i32:%0
122 %1 = add i32 %0, %0 ; yields i32:%1
123 %result = add i32 %1, %1
125 This last way of multiplying ``%X`` by 8 illustrates several important
126 lexical features of LLVM:
128 #. Comments are delimited with a '``;``' and go until the end of line.
129 #. Unnamed temporaries are created when the result of a computation is
130 not assigned to a named value.
131 #. Unnamed temporaries are numbered sequentially (using a per-function
132 incrementing counter, starting with 0). Note that basic blocks and unnamed
133 function parameters are included in this numbering. For example, if the
134 entry basic block is not given a label name and all function parameters are
135 named, then it will get number 0.
137 It also shows a convention that we follow in this document. When
138 demonstrating instructions, we will follow an instruction with a comment
139 that defines the type and name of value produced.
147 LLVM programs are composed of ``Module``'s, each of which is a
148 translation unit of the input programs. Each module consists of
149 functions, global variables, and symbol table entries. Modules may be
150 combined together with the LLVM linker, which merges function (and
151 global variable) definitions, resolves forward declarations, and merges
152 symbol table entries. Here is an example of the "hello world" module:
156 ; Declare the string constant as a global constant.
157 @.str = private unnamed_addr constant [13 x i8] c"hello world\0A\00"
159 ; External declaration of the puts function
160 declare i32 @puts(i8* nocapture) nounwind
162 ; Definition of main function
163 define i32 @main() { ; i32()*
164 ; Convert [13 x i8]* to i8 *...
165 %cast210 = getelementptr [13 x i8], [13 x i8]* @.str, i64 0, i64 0
167 ; Call puts function to write out the string to stdout.
168 call i32 @puts(i8* %cast210)
173 !0 = !{i32 42, null, !"string"}
176 This example is made up of a :ref:`global variable <globalvars>` named
177 "``.str``", an external declaration of the "``puts``" function, a
178 :ref:`function definition <functionstructure>` for "``main``" and
179 :ref:`named metadata <namedmetadatastructure>` "``foo``".
181 In general, a module is made up of a list of global values (where both
182 functions and global variables are global values). Global values are
183 represented by a pointer to a memory location (in this case, a pointer
184 to an array of char, and a pointer to a function), and have one of the
185 following :ref:`linkage types <linkage>`.
192 All Global Variables and Functions have one of the following types of
196 Global values with "``private``" linkage are only directly
197 accessible by objects in the current module. In particular, linking
198 code into a module with an private global value may cause the
199 private to be renamed as necessary to avoid collisions. Because the
200 symbol is private to the module, all references can be updated. This
201 doesn't show up in any symbol table in the object file.
203 Similar to private, but the value shows as a local symbol
204 (``STB_LOCAL`` in the case of ELF) in the object file. This
205 corresponds to the notion of the '``static``' keyword in C.
206 ``available_externally``
207 Globals with "``available_externally``" linkage are never emitted
208 into the object file corresponding to the LLVM module. They exist to
209 allow inlining and other optimizations to take place given knowledge
210 of the definition of the global, which is known to be somewhere
211 outside the module. Globals with ``available_externally`` linkage
212 are allowed to be discarded at will, and are otherwise the same as
213 ``linkonce_odr``. This linkage type is only allowed on definitions,
216 Globals with "``linkonce``" linkage are merged with other globals of
217 the same name when linkage occurs. This can be used to implement
218 some forms of inline functions, templates, or other code which must
219 be generated in each translation unit that uses it, but where the
220 body may be overridden with a more definitive definition later.
221 Unreferenced ``linkonce`` globals are allowed to be discarded. Note
222 that ``linkonce`` linkage does not actually allow the optimizer to
223 inline the body of this function into callers because it doesn't
224 know if this definition of the function is the definitive definition
225 within the program or whether it will be overridden by a stronger
226 definition. To enable inlining and other optimizations, use
227 "``linkonce_odr``" linkage.
229 "``weak``" linkage has the same merging semantics as ``linkonce``
230 linkage, except that unreferenced globals with ``weak`` linkage may
231 not be discarded. This is used for globals that are declared "weak"
234 "``common``" linkage is most similar to "``weak``" linkage, but they
235 are used for tentative definitions in C, such as "``int X;``" at
236 global scope. Symbols with "``common``" linkage are merged in the
237 same way as ``weak symbols``, and they may not be deleted if
238 unreferenced. ``common`` symbols may not have an explicit section,
239 must have a zero initializer, and may not be marked
240 ':ref:`constant <globalvars>`'. Functions and aliases may not have
243 .. _linkage_appending:
246 "``appending``" linkage may only be applied to global variables of
247 pointer to array type. When two global variables with appending
248 linkage are linked together, the two global arrays are appended
249 together. This is the LLVM, typesafe, equivalent of having the
250 system linker append together "sections" with identical names when
253 The semantics of this linkage follow the ELF object file model: the
254 symbol is weak until linked, if not linked, the symbol becomes null
255 instead of being an undefined reference.
256 ``linkonce_odr``, ``weak_odr``
257 Some languages allow differing globals to be merged, such as two
258 functions with different semantics. Other languages, such as
259 ``C++``, ensure that only equivalent globals are ever merged (the
260 "one definition rule" --- "ODR"). Such languages can use the
261 ``linkonce_odr`` and ``weak_odr`` linkage types to indicate that the
262 global will only be merged with equivalent globals. These linkage
263 types are otherwise the same as their non-``odr`` versions.
265 If none of the above identifiers are used, the global is externally
266 visible, meaning that it participates in linkage and can be used to
267 resolve external symbol references.
269 It is illegal for a function *declaration* to have any linkage type
270 other than ``external`` or ``extern_weak``.
277 LLVM :ref:`functions <functionstructure>`, :ref:`calls <i_call>` and
278 :ref:`invokes <i_invoke>` can all have an optional calling convention
279 specified for the call. The calling convention of any pair of dynamic
280 caller/callee must match, or the behavior of the program is undefined.
281 The following calling conventions are supported by LLVM, and more may be
284 "``ccc``" - The C calling convention
285 This calling convention (the default if no other calling convention
286 is specified) matches the target C calling conventions. This calling
287 convention supports varargs function calls and tolerates some
288 mismatch in the declared prototype and implemented declaration of
289 the function (as does normal C).
290 "``fastcc``" - The fast calling convention
291 This calling convention attempts to make calls as fast as possible
292 (e.g. by passing things in registers). This calling convention
293 allows the target to use whatever tricks it wants to produce fast
294 code for the target, without having to conform to an externally
295 specified ABI (Application Binary Interface). `Tail calls can only
296 be optimized when this, the GHC or the HiPE convention is
297 used. <CodeGenerator.html#id80>`_ This calling convention does not
298 support varargs and requires the prototype of all callees to exactly
299 match the prototype of the function definition.
300 "``coldcc``" - The cold calling convention
301 This calling convention attempts to make code in the caller as
302 efficient as possible under the assumption that the call is not
303 commonly executed. As such, these calls often preserve all registers
304 so that the call does not break any live ranges in the caller side.
305 This calling convention does not support varargs and requires the
306 prototype of all callees to exactly match the prototype of the
307 function definition. Furthermore the inliner doesn't consider such function
309 "``cc 10``" - GHC convention
310 This calling convention has been implemented specifically for use by
311 the `Glasgow Haskell Compiler (GHC) <http://www.haskell.org/ghc>`_.
312 It passes everything in registers, going to extremes to achieve this
313 by disabling callee save registers. This calling convention should
314 not be used lightly but only for specific situations such as an
315 alternative to the *register pinning* performance technique often
316 used when implementing functional programming languages. At the
317 moment only X86 supports this convention and it has the following
320 - On *X86-32* only supports up to 4 bit type parameters. No
321 floating point types are supported.
322 - On *X86-64* only supports up to 10 bit type parameters and 6
323 floating point parameters.
325 This calling convention supports `tail call
326 optimization <CodeGenerator.html#id80>`_ but requires both the
327 caller and callee are using it.
328 "``cc 11``" - The HiPE calling convention
329 This calling convention has been implemented specifically for use by
330 the `High-Performance Erlang
331 (HiPE) <http://www.it.uu.se/research/group/hipe/>`_ compiler, *the*
332 native code compiler of the `Ericsson's Open Source Erlang/OTP
333 system <http://www.erlang.org/download.shtml>`_. It uses more
334 registers for argument passing than the ordinary C calling
335 convention and defines no callee-saved registers. The calling
336 convention properly supports `tail call
337 optimization <CodeGenerator.html#id80>`_ but requires that both the
338 caller and the callee use it. It uses a *register pinning*
339 mechanism, similar to GHC's convention, for keeping frequently
340 accessed runtime components pinned to specific hardware registers.
341 At the moment only X86 supports this convention (both 32 and 64
343 "``webkit_jscc``" - WebKit's JavaScript calling convention
344 This calling convention has been implemented for `WebKit FTL JIT
345 <https://trac.webkit.org/wiki/FTLJIT>`_. It passes arguments on the
346 stack right to left (as cdecl does), and returns a value in the
347 platform's customary return register.
348 "``anyregcc``" - Dynamic calling convention for code patching
349 This is a special convention that supports patching an arbitrary code
350 sequence in place of a call site. This convention forces the call
351 arguments into registers but allows them to be dynamically
352 allocated. This can currently only be used with calls to
353 llvm.experimental.patchpoint because only this intrinsic records
354 the location of its arguments in a side table. See :doc:`StackMaps`.
355 "``preserve_mostcc``" - The `PreserveMost` calling convention
356 This calling convention attempts to make the code in the caller as
357 unintrusive as possible. This convention behaves identically to the `C`
358 calling convention on how arguments and return values are passed, but it
359 uses a different set of caller/callee-saved registers. This alleviates the
360 burden of saving and recovering a large register set before and after the
361 call in the caller. If the arguments are passed in callee-saved registers,
362 then they will be preserved by the callee across the call. This doesn't
363 apply for values returned in callee-saved registers.
365 - On X86-64 the callee preserves all general purpose registers, except for
366 R11. R11 can be used as a scratch register. Floating-point registers
367 (XMMs/YMMs) are not preserved and need to be saved by the caller.
369 The idea behind this convention is to support calls to runtime functions
370 that have a hot path and a cold path. The hot path is usually a small piece
371 of code that doesn't use many registers. The cold path might need to call out to
372 another function and therefore only needs to preserve the caller-saved
373 registers, which haven't already been saved by the caller. The
374 `PreserveMost` calling convention is very similar to the `cold` calling
375 convention in terms of caller/callee-saved registers, but they are used for
376 different types of function calls. `coldcc` is for function calls that are
377 rarely executed, whereas `preserve_mostcc` function calls are intended to be
378 on the hot path and definitely executed a lot. Furthermore `preserve_mostcc`
379 doesn't prevent the inliner from inlining the function call.
381 This calling convention will be used by a future version of the ObjectiveC
382 runtime and should therefore still be considered experimental at this time.
383 Although this convention was created to optimize certain runtime calls to
384 the ObjectiveC runtime, it is not limited to this runtime and might be used
385 by other runtimes in the future too. The current implementation only
386 supports X86-64, but the intention is to support more architectures in the
388 "``preserve_allcc``" - The `PreserveAll` calling convention
389 This calling convention attempts to make the code in the caller even less
390 intrusive than the `PreserveMost` calling convention. This calling
391 convention also behaves identical to the `C` calling convention on how
392 arguments and return values are passed, but it uses a different set of
393 caller/callee-saved registers. This removes the burden of saving and
394 recovering a large register set before and after the call in the caller. If
395 the arguments are passed in callee-saved registers, then they will be
396 preserved by the callee across the call. This doesn't apply for values
397 returned in callee-saved registers.
399 - On X86-64 the callee preserves all general purpose registers, except for
400 R11. R11 can be used as a scratch register. Furthermore it also preserves
401 all floating-point registers (XMMs/YMMs).
403 The idea behind this convention is to support calls to runtime functions
404 that don't need to call out to any other functions.
406 This calling convention, like the `PreserveMost` calling convention, will be
407 used by a future version of the ObjectiveC runtime and should be considered
408 experimental at this time.
409 "``cc <n>``" - Numbered convention
410 Any calling convention may be specified by number, allowing
411 target-specific calling conventions to be used. Target specific
412 calling conventions start at 64.
414 More calling conventions can be added/defined on an as-needed basis, to
415 support Pascal conventions or any other well-known target-independent
418 .. _visibilitystyles:
423 All Global Variables and Functions have one of the following visibility
426 "``default``" - Default style
427 On targets that use the ELF object file format, default visibility
428 means that the declaration is visible to other modules and, in
429 shared libraries, means that the declared entity may be overridden.
430 On Darwin, default visibility means that the declaration is visible
431 to other modules. Default visibility corresponds to "external
432 linkage" in the language.
433 "``hidden``" - Hidden style
434 Two declarations of an object with hidden visibility refer to the
435 same object if they are in the same shared object. Usually, hidden
436 visibility indicates that the symbol will not be placed into the
437 dynamic symbol table, so no other module (executable or shared
438 library) can reference it directly.
439 "``protected``" - Protected style
440 On ELF, protected visibility indicates that the symbol will be
441 placed in the dynamic symbol table, but that references within the
442 defining module will bind to the local symbol. That is, the symbol
443 cannot be overridden by another module.
445 A symbol with ``internal`` or ``private`` linkage must have ``default``
453 All Global Variables, Functions and Aliases can have one of the following
457 "``dllimport``" causes the compiler to reference a function or variable via
458 a global pointer to a pointer that is set up by the DLL exporting the
459 symbol. On Microsoft Windows targets, the pointer name is formed by
460 combining ``__imp_`` and the function or variable name.
462 "``dllexport``" causes the compiler to provide a global pointer to a pointer
463 in a DLL, so that it can be referenced with the ``dllimport`` attribute. On
464 Microsoft Windows targets, the pointer name is formed by combining
465 ``__imp_`` and the function or variable name. Since this storage class
466 exists for defining a dll interface, the compiler, assembler and linker know
467 it is externally referenced and must refrain from deleting the symbol.
471 Thread Local Storage Models
472 ---------------------------
474 A variable may be defined as ``thread_local``, which means that it will
475 not be shared by threads (each thread will have a separated copy of the
476 variable). Not all targets support thread-local variables. Optionally, a
477 TLS model may be specified:
480 For variables that are only used within the current shared library.
482 For variables in modules that will not be loaded dynamically.
484 For variables defined in the executable and only used within it.
486 If no explicit model is given, the "general dynamic" model is used.
488 The models correspond to the ELF TLS models; see `ELF Handling For
489 Thread-Local Storage <http://people.redhat.com/drepper/tls.pdf>`_ for
490 more information on under which circumstances the different models may
491 be used. The target may choose a different TLS model if the specified
492 model is not supported, or if a better choice of model can be made.
494 A model can also be specified in a alias, but then it only governs how
495 the alias is accessed. It will not have any effect in the aliasee.
502 LLVM IR allows you to specify both "identified" and "literal" :ref:`structure
503 types <t_struct>`. Literal types are uniqued structurally, but identified types
504 are never uniqued. An :ref:`opaque structural type <t_opaque>` can also be used
505 to forward declare a type that is not yet available.
507 An example of a identified structure specification is:
511 %mytype = type { %mytype*, i32 }
513 Prior to the LLVM 3.0 release, identified types were structurally uniqued. Only
514 literal types are uniqued in recent versions of LLVM.
521 Global variables define regions of memory allocated at compilation time
524 Global variable definitions must be initialized.
526 Global variables in other translation units can also be declared, in which
527 case they don't have an initializer.
529 Either global variable definitions or declarations may have an explicit section
530 to be placed in and may have an optional explicit alignment specified.
532 A variable may be defined as a global ``constant``, which indicates that
533 the contents of the variable will **never** be modified (enabling better
534 optimization, allowing the global data to be placed in the read-only
535 section of an executable, etc). Note that variables that need runtime
536 initialization cannot be marked ``constant`` as there is a store to the
539 LLVM explicitly allows *declarations* of global variables to be marked
540 constant, even if the final definition of the global is not. This
541 capability can be used to enable slightly better optimization of the
542 program, but requires the language definition to guarantee that
543 optimizations based on the 'constantness' are valid for the translation
544 units that do not include the definition.
546 As SSA values, global variables define pointer values that are in scope
547 (i.e. they dominate) all basic blocks in the program. Global variables
548 always define a pointer to their "content" type because they describe a
549 region of memory, and all memory objects in LLVM are accessed through
552 Global variables can be marked with ``unnamed_addr`` which indicates
553 that the address is not significant, only the content. Constants marked
554 like this can be merged with other constants if they have the same
555 initializer. Note that a constant with significant address *can* be
556 merged with a ``unnamed_addr`` constant, the result being a constant
557 whose address is significant.
559 A global variable may be declared to reside in a target-specific
560 numbered address space. For targets that support them, address spaces
561 may affect how optimizations are performed and/or what target
562 instructions are used to access the variable. The default address space
563 is zero. The address space qualifier must precede any other attributes.
565 LLVM allows an explicit section to be specified for globals. If the
566 target supports it, it will emit globals to the section specified.
567 Additionally, the global can placed in a comdat if the target has the necessary
570 By default, global initializers are optimized by assuming that global
571 variables defined within the module are not modified from their
572 initial values before the start of the global initializer. This is
573 true even for variables potentially accessible from outside the
574 module, including those with external linkage or appearing in
575 ``@llvm.used`` or dllexported variables. This assumption may be suppressed
576 by marking the variable with ``externally_initialized``.
578 An explicit alignment may be specified for a global, which must be a
579 power of 2. If not present, or if the alignment is set to zero, the
580 alignment of the global is set by the target to whatever it feels
581 convenient. If an explicit alignment is specified, the global is forced
582 to have exactly that alignment. Targets and optimizers are not allowed
583 to over-align the global if the global has an assigned section. In this
584 case, the extra alignment could be observable: for example, code could
585 assume that the globals are densely packed in their section and try to
586 iterate over them as an array, alignment padding would break this
587 iteration. The maximum alignment is ``1 << 29``.
589 Globals can also have a :ref:`DLL storage class <dllstorageclass>`.
591 Variables and aliases can have a
592 :ref:`Thread Local Storage Model <tls_model>`.
596 [@<GlobalVarName> =] [Linkage] [Visibility] [DLLStorageClass] [ThreadLocal]
597 [unnamed_addr] [AddrSpace] [ExternallyInitialized]
598 <global | constant> <Type> [<InitializerConstant>]
599 [, section "name"] [, comdat [($name)]]
600 [, align <Alignment>]
602 For example, the following defines a global in a numbered address space
603 with an initializer, section, and alignment:
607 @G = addrspace(5) constant float 1.0, section "foo", align 4
609 The following example just declares a global variable
613 @G = external global i32
615 The following example defines a thread-local global with the
616 ``initialexec`` TLS model:
620 @G = thread_local(initialexec) global i32 0, align 4
622 .. _functionstructure:
627 LLVM function definitions consist of the "``define``" keyword, an
628 optional :ref:`linkage type <linkage>`, an optional :ref:`visibility
629 style <visibility>`, an optional :ref:`DLL storage class <dllstorageclass>`,
630 an optional :ref:`calling convention <callingconv>`,
631 an optional ``unnamed_addr`` attribute, a return type, an optional
632 :ref:`parameter attribute <paramattrs>` for the return type, a function
633 name, a (possibly empty) argument list (each with optional :ref:`parameter
634 attributes <paramattrs>`), optional :ref:`function attributes <fnattrs>`,
635 an optional section, an optional alignment,
636 an optional :ref:`comdat <langref_comdats>`,
637 an optional :ref:`garbage collector name <gc>`, an optional :ref:`prefix <prefixdata>`,
638 an optional :ref:`prologue <prologuedata>`, an opening
639 curly brace, a list of basic blocks, and a closing curly brace.
641 LLVM function declarations consist of the "``declare``" keyword, an
642 optional :ref:`linkage type <linkage>`, an optional :ref:`visibility
643 style <visibility>`, an optional :ref:`DLL storage class <dllstorageclass>`,
644 an optional :ref:`calling convention <callingconv>`,
645 an optional ``unnamed_addr`` attribute, a return type, an optional
646 :ref:`parameter attribute <paramattrs>` for the return type, a function
647 name, a possibly empty list of arguments, an optional alignment, an optional
648 :ref:`garbage collector name <gc>`, an optional :ref:`prefix <prefixdata>`,
649 and an optional :ref:`prologue <prologuedata>`.
651 A function definition contains a list of basic blocks, forming the CFG (Control
652 Flow Graph) for the function. Each basic block may optionally start with a label
653 (giving the basic block a symbol table entry), contains a list of instructions,
654 and ends with a :ref:`terminator <terminators>` instruction (such as a branch or
655 function return). If an explicit label is not provided, a block is assigned an
656 implicit numbered label, using the next value from the same counter as used for
657 unnamed temporaries (:ref:`see above<identifiers>`). For example, if a function
658 entry block does not have an explicit label, it will be assigned label "%0",
659 then the first unnamed temporary in that block will be "%1", etc.
661 The first basic block in a function is special in two ways: it is
662 immediately executed on entrance to the function, and it is not allowed
663 to have predecessor basic blocks (i.e. there can not be any branches to
664 the entry block of a function). Because the block can have no
665 predecessors, it also cannot have any :ref:`PHI nodes <i_phi>`.
667 LLVM allows an explicit section to be specified for functions. If the
668 target supports it, it will emit functions to the section specified.
669 Additionally, the function can be placed in a COMDAT.
671 An explicit alignment may be specified for a function. If not present,
672 or if the alignment is set to zero, the alignment of the function is set
673 by the target to whatever it feels convenient. If an explicit alignment
674 is specified, the function is forced to have at least that much
675 alignment. All alignments must be a power of 2.
677 If the ``unnamed_addr`` attribute is given, the address is known to not
678 be significant and two identical functions can be merged.
682 define [linkage] [visibility] [DLLStorageClass]
684 <ResultType> @<FunctionName> ([argument list])
685 [unnamed_addr] [fn Attrs] [section "name"] [comdat [($name)]]
686 [align N] [gc] [prefix Constant] [prologue Constant] { ... }
688 The argument list is a comma seperated sequence of arguments where each
689 argument is of the following form
693 <type> [parameter Attrs] [name]
701 Aliases, unlike function or variables, don't create any new data. They
702 are just a new symbol and metadata for an existing position.
704 Aliases have a name and an aliasee that is either a global value or a
707 Aliases may have an optional :ref:`linkage type <linkage>`, an optional
708 :ref:`visibility style <visibility>`, an optional :ref:`DLL storage class
709 <dllstorageclass>` and an optional :ref:`tls model <tls_model>`.
713 @<Name> = [Linkage] [Visibility] [DLLStorageClass] [ThreadLocal] [unnamed_addr] alias <AliaseeTy> @<Aliasee>
715 The linkage must be one of ``private``, ``internal``, ``linkonce``, ``weak``,
716 ``linkonce_odr``, ``weak_odr``, ``external``. Note that some system linkers
717 might not correctly handle dropping a weak symbol that is aliased.
719 Aliases that are not ``unnamed_addr`` are guaranteed to have the same address as
720 the aliasee expression. ``unnamed_addr`` ones are only guaranteed to point
723 Since aliases are only a second name, some restrictions apply, of which
724 some can only be checked when producing an object file:
726 * The expression defining the aliasee must be computable at assembly
727 time. Since it is just a name, no relocations can be used.
729 * No alias in the expression can be weak as the possibility of the
730 intermediate alias being overridden cannot be represented in an
733 * No global value in the expression can be a declaration, since that
734 would require a relocation, which is not possible.
741 Comdat IR provides access to COFF and ELF object file COMDAT functionality.
743 Comdats have a name which represents the COMDAT key. All global objects that
744 specify this key will only end up in the final object file if the linker chooses
745 that key over some other key. Aliases are placed in the same COMDAT that their
746 aliasee computes to, if any.
748 Comdats have a selection kind to provide input on how the linker should
749 choose between keys in two different object files.
753 $<Name> = comdat SelectionKind
755 The selection kind must be one of the following:
758 The linker may choose any COMDAT key, the choice is arbitrary.
760 The linker may choose any COMDAT key but the sections must contain the
763 The linker will choose the section containing the largest COMDAT key.
765 The linker requires that only section with this COMDAT key exist.
767 The linker may choose any COMDAT key but the sections must contain the
770 Note that the Mach-O platform doesn't support COMDATs and ELF only supports
771 ``any`` as a selection kind.
773 Here is an example of a COMDAT group where a function will only be selected if
774 the COMDAT key's section is the largest:
778 $foo = comdat largest
779 @foo = global i32 2, comdat($foo)
781 define void @bar() comdat($foo) {
785 As a syntactic sugar the ``$name`` can be omitted if the name is the same as
791 @foo = global i32 2, comdat
794 In a COFF object file, this will create a COMDAT section with selection kind
795 ``IMAGE_COMDAT_SELECT_LARGEST`` containing the contents of the ``@foo`` symbol
796 and another COMDAT section with selection kind
797 ``IMAGE_COMDAT_SELECT_ASSOCIATIVE`` which is associated with the first COMDAT
798 section and contains the contents of the ``@bar`` symbol.
800 There are some restrictions on the properties of the global object.
801 It, or an alias to it, must have the same name as the COMDAT group when
803 The contents and size of this object may be used during link-time to determine
804 which COMDAT groups get selected depending on the selection kind.
805 Because the name of the object must match the name of the COMDAT group, the
806 linkage of the global object must not be local; local symbols can get renamed
807 if a collision occurs in the symbol table.
809 The combined use of COMDATS and section attributes may yield surprising results.
816 @g1 = global i32 42, section "sec", comdat($foo)
817 @g2 = global i32 42, section "sec", comdat($bar)
819 From the object file perspective, this requires the creation of two sections
820 with the same name. This is necessary because both globals belong to different
821 COMDAT groups and COMDATs, at the object file level, are represented by
824 Note that certain IR constructs like global variables and functions may create
825 COMDATs in the object file in addition to any which are specified using COMDAT
826 IR. This arises, for example, when a global variable has linkonce_odr linkage.
828 .. _namedmetadatastructure:
833 Named metadata is a collection of metadata. :ref:`Metadata
834 nodes <metadata>` (but not metadata strings) are the only valid
835 operands for a named metadata.
839 ; Some unnamed metadata nodes, which are referenced by the named metadata.
844 !name = !{!0, !1, !2}
851 The return type and each parameter of a function type may have a set of
852 *parameter attributes* associated with them. Parameter attributes are
853 used to communicate additional information about the result or
854 parameters of a function. Parameter attributes are considered to be part
855 of the function, not of the function type, so functions with different
856 parameter attributes can have the same function type.
858 Parameter attributes are simple keywords that follow the type specified.
859 If multiple parameter attributes are needed, they are space separated.
864 declare i32 @printf(i8* noalias nocapture, ...)
865 declare i32 @atoi(i8 zeroext)
866 declare signext i8 @returns_signed_char()
868 Note that any attributes for the function result (``nounwind``,
869 ``readonly``) come immediately after the argument list.
871 Currently, only the following parameter attributes are defined:
874 This indicates to the code generator that the parameter or return
875 value should be zero-extended to the extent required by the target's
876 ABI (which is usually 32-bits, but is 8-bits for a i1 on x86-64) by
877 the caller (for a parameter) or the callee (for a return value).
879 This indicates to the code generator that the parameter or return
880 value should be sign-extended to the extent required by the target's
881 ABI (which is usually 32-bits) by the caller (for a parameter) or
882 the callee (for a return value).
884 This indicates that this parameter or return value should be treated
885 in a special target-dependent fashion during while emitting code for
886 a function call or return (usually, by putting it in a register as
887 opposed to memory, though some targets use it to distinguish between
888 two different kinds of registers). Use of this attribute is
891 This indicates that the pointer parameter should really be passed by
892 value to the function. The attribute implies that a hidden copy of
893 the pointee is made between the caller and the callee, so the callee
894 is unable to modify the value in the caller. This attribute is only
895 valid on LLVM pointer arguments. It is generally used to pass
896 structs and arrays by value, but is also valid on pointers to
897 scalars. The copy is considered to belong to the caller not the
898 callee (for example, ``readonly`` functions should not write to
899 ``byval`` parameters). This is not a valid attribute for return
902 The byval attribute also supports specifying an alignment with the
903 align attribute. It indicates the alignment of the stack slot to
904 form and the known alignment of the pointer specified to the call
905 site. If the alignment is not specified, then the code generator
906 makes a target-specific assumption.
912 The ``inalloca`` argument attribute allows the caller to take the
913 address of outgoing stack arguments. An ``inalloca`` argument must
914 be a pointer to stack memory produced by an ``alloca`` instruction.
915 The alloca, or argument allocation, must also be tagged with the
916 inalloca keyword. Only the last argument may have the ``inalloca``
917 attribute, and that argument is guaranteed to be passed in memory.
919 An argument allocation may be used by a call at most once because
920 the call may deallocate it. The ``inalloca`` attribute cannot be
921 used in conjunction with other attributes that affect argument
922 storage, like ``inreg``, ``nest``, ``sret``, or ``byval``. The
923 ``inalloca`` attribute also disables LLVM's implicit lowering of
924 large aggregate return values, which means that frontend authors
925 must lower them with ``sret`` pointers.
927 When the call site is reached, the argument allocation must have
928 been the most recent stack allocation that is still live, or the
929 results are undefined. It is possible to allocate additional stack
930 space after an argument allocation and before its call site, but it
931 must be cleared off with :ref:`llvm.stackrestore
934 See :doc:`InAlloca` for more information on how to use this
938 This indicates that the pointer parameter specifies the address of a
939 structure that is the return value of the function in the source
940 program. This pointer must be guaranteed by the caller to be valid:
941 loads and stores to the structure may be assumed by the callee
942 not to trap and to be properly aligned. This may only be applied to
943 the first parameter. This is not a valid attribute for return
947 This indicates that the pointer value may be assumed by the optimizer to
948 have the specified alignment.
950 Note that this attribute has additional semantics when combined with the
956 This indicates that objects accessed via pointer values
957 :ref:`based <pointeraliasing>` on the argument or return value are not also
958 accessed, during the execution of the function, via pointer values not
959 *based* on the argument or return value. The attribute on a return value
960 also has additional semantics described below. The caller shares the
961 responsibility with the callee for ensuring that these requirements are met.
962 For further details, please see the discussion of the NoAlias response in
963 :ref:`alias analysis <Must, May, or No>`.
965 Note that this definition of ``noalias`` is intentionally similar
966 to the definition of ``restrict`` in C99 for function arguments.
968 For function return values, C99's ``restrict`` is not meaningful,
969 while LLVM's ``noalias`` is. Furthermore, the semantics of the ``noalias``
970 attribute on return values are stronger than the semantics of the attribute
971 when used on function arguments. On function return values, the ``noalias``
972 attribute indicates that the function acts like a system memory allocation
973 function, returning a pointer to allocated storage disjoint from the
974 storage for any other object accessible to the caller.
977 This indicates that the callee does not make any copies of the
978 pointer that outlive the callee itself. This is not a valid
979 attribute for return values.
984 This indicates that the pointer parameter can be excised using the
985 :ref:`trampoline intrinsics <int_trampoline>`. This is not a valid
986 attribute for return values and can only be applied to one parameter.
989 This indicates that the function always returns the argument as its return
990 value. This is an optimization hint to the code generator when generating
991 the caller, allowing tail call optimization and omission of register saves
992 and restores in some cases; it is not checked or enforced when generating
993 the callee. The parameter and the function return type must be valid
994 operands for the :ref:`bitcast instruction <i_bitcast>`. This is not a
995 valid attribute for return values and can only be applied to one parameter.
998 This indicates that the parameter or return pointer is not null. This
999 attribute may only be applied to pointer typed parameters. This is not
1000 checked or enforced by LLVM, the caller must ensure that the pointer
1001 passed in is non-null, or the callee must ensure that the returned pointer
1004 ``dereferenceable(<n>)``
1005 This indicates that the parameter or return pointer is dereferenceable. This
1006 attribute may only be applied to pointer typed parameters. A pointer that
1007 is dereferenceable can be loaded from speculatively without a risk of
1008 trapping. The number of bytes known to be dereferenceable must be provided
1009 in parentheses. It is legal for the number of bytes to be less than the
1010 size of the pointee type. The ``nonnull`` attribute does not imply
1011 dereferenceability (consider a pointer to one element past the end of an
1012 array), however ``dereferenceable(<n>)`` does imply ``nonnull`` in
1013 ``addrspace(0)`` (which is the default address space).
1015 ``dereferenceable_or_null(<n>)``
1016 This indicates that the parameter or return value isn't both
1017 non-null and non-dereferenceable (up to ``<n>`` bytes) at the same
1018 time. All non-null pointers tagged with
1019 ``dereferenceable_or_null(<n>)`` are ``dereferenceable(<n>)``.
1020 For address space 0 ``dereferenceable_or_null(<n>)`` implies that
1021 a pointer is exactly one of ``dereferenceable(<n>)`` or ``null``,
1022 and in other address spaces ``dereferenceable_or_null(<n>)``
1023 implies that a pointer is at least one of ``dereferenceable(<n>)``
1024 or ``null`` (i.e. it may be both ``null`` and
1025 ``dereferenceable(<n>)``). This attribute may only be applied to
1026 pointer typed parameters.
1030 Garbage Collector Strategy Names
1031 --------------------------------
1033 Each function may specify a garbage collector strategy name, which is simply a
1036 .. code-block:: llvm
1038 define void @f() gc "name" { ... }
1040 The supported values of *name* includes those :ref:`built in to LLVM
1041 <builtin-gc-strategies>` and any provided by loaded plugins. Specifying a GC
1042 strategy will cause the compiler to alter its output in order to support the
1043 named garbage collection algorithm. Note that LLVM itself does not contain a
1044 garbage collector, this functionality is restricted to generating machine code
1045 which can interoperate with a collector provided externally.
1052 Prefix data is data associated with a function which the code
1053 generator will emit immediately before the function's entrypoint.
1054 The purpose of this feature is to allow frontends to associate
1055 language-specific runtime metadata with specific functions and make it
1056 available through the function pointer while still allowing the
1057 function pointer to be called.
1059 To access the data for a given function, a program may bitcast the
1060 function pointer to a pointer to the constant's type and dereference
1061 index -1. This implies that the IR symbol points just past the end of
1062 the prefix data. For instance, take the example of a function annotated
1063 with a single ``i32``,
1065 .. code-block:: llvm
1067 define void @f() prefix i32 123 { ... }
1069 The prefix data can be referenced as,
1071 .. code-block:: llvm
1073 %0 = bitcast void* () @f to i32*
1074 %a = getelementptr inbounds i32, i32* %0, i32 -1
1075 %b = load i32, i32* %a
1077 Prefix data is laid out as if it were an initializer for a global variable
1078 of the prefix data's type. The function will be placed such that the
1079 beginning of the prefix data is aligned. This means that if the size
1080 of the prefix data is not a multiple of the alignment size, the
1081 function's entrypoint will not be aligned. If alignment of the
1082 function's entrypoint is desired, padding must be added to the prefix
1085 A function may have prefix data but no body. This has similar semantics
1086 to the ``available_externally`` linkage in that the data may be used by the
1087 optimizers but will not be emitted in the object file.
1094 The ``prologue`` attribute allows arbitrary code (encoded as bytes) to
1095 be inserted prior to the function body. This can be used for enabling
1096 function hot-patching and instrumentation.
1098 To maintain the semantics of ordinary function calls, the prologue data must
1099 have a particular format. Specifically, it must begin with a sequence of
1100 bytes which decode to a sequence of machine instructions, valid for the
1101 module's target, which transfer control to the point immediately succeeding
1102 the prologue data, without performing any other visible action. This allows
1103 the inliner and other passes to reason about the semantics of the function
1104 definition without needing to reason about the prologue data. Obviously this
1105 makes the format of the prologue data highly target dependent.
1107 A trivial example of valid prologue data for the x86 architecture is ``i8 144``,
1108 which encodes the ``nop`` instruction:
1110 .. code-block:: llvm
1112 define void @f() prologue i8 144 { ... }
1114 Generally prologue data can be formed by encoding a relative branch instruction
1115 which skips the metadata, as in this example of valid prologue data for the
1116 x86_64 architecture, where the first two bytes encode ``jmp .+10``:
1118 .. code-block:: llvm
1120 %0 = type <{ i8, i8, i8* }>
1122 define void @f() prologue %0 <{ i8 235, i8 8, i8* @md}> { ... }
1124 A function may have prologue data but no body. This has similar semantics
1125 to the ``available_externally`` linkage in that the data may be used by the
1126 optimizers but will not be emitted in the object file.
1133 Attribute groups are groups of attributes that are referenced by objects within
1134 the IR. They are important for keeping ``.ll`` files readable, because a lot of
1135 functions will use the same set of attributes. In the degenerative case of a
1136 ``.ll`` file that corresponds to a single ``.c`` file, the single attribute
1137 group will capture the important command line flags used to build that file.
1139 An attribute group is a module-level object. To use an attribute group, an
1140 object references the attribute group's ID (e.g. ``#37``). An object may refer
1141 to more than one attribute group. In that situation, the attributes from the
1142 different groups are merged.
1144 Here is an example of attribute groups for a function that should always be
1145 inlined, has a stack alignment of 4, and which shouldn't use SSE instructions:
1147 .. code-block:: llvm
1149 ; Target-independent attributes:
1150 attributes #0 = { alwaysinline alignstack=4 }
1152 ; Target-dependent attributes:
1153 attributes #1 = { "no-sse" }
1155 ; Function @f has attributes: alwaysinline, alignstack=4, and "no-sse".
1156 define void @f() #0 #1 { ... }
1163 Function attributes are set to communicate additional information about
1164 a function. Function attributes are considered to be part of the
1165 function, not of the function type, so functions with different function
1166 attributes can have the same function type.
1168 Function attributes are simple keywords that follow the type specified.
1169 If multiple attributes are needed, they are space separated. For
1172 .. code-block:: llvm
1174 define void @f() noinline { ... }
1175 define void @f() alwaysinline { ... }
1176 define void @f() alwaysinline optsize { ... }
1177 define void @f() optsize { ... }
1180 This attribute indicates that, when emitting the prologue and
1181 epilogue, the backend should forcibly align the stack pointer.
1182 Specify the desired alignment, which must be a power of two, in
1185 This attribute indicates that the inliner should attempt to inline
1186 this function into callers whenever possible, ignoring any active
1187 inlining size threshold for this caller.
1189 This indicates that the callee function at a call site should be
1190 recognized as a built-in function, even though the function's declaration
1191 uses the ``nobuiltin`` attribute. This is only valid at call sites for
1192 direct calls to functions that are declared with the ``nobuiltin``
1195 This attribute indicates that this function is rarely called. When
1196 computing edge weights, basic blocks post-dominated by a cold
1197 function call are also considered to be cold; and, thus, given low
1200 This attribute indicates that the callee is dependent on a convergent
1201 thread execution pattern under certain parallel execution models.
1202 Transformations that are execution model agnostic may only move or
1203 tranform this call if the final location is control equivalent to its
1204 original position in the program, where control equivalence is defined as
1205 A dominates B and B post-dominates A, or vice versa.
1207 This attribute indicates that the source code contained a hint that
1208 inlining this function is desirable (such as the "inline" keyword in
1209 C/C++). It is just a hint; it imposes no requirements on the
1212 This attribute indicates that the function should be added to a
1213 jump-instruction table at code-generation time, and that all address-taken
1214 references to this function should be replaced with a reference to the
1215 appropriate jump-instruction-table function pointer. Note that this creates
1216 a new pointer for the original function, which means that code that depends
1217 on function-pointer identity can break. So, any function annotated with
1218 ``jumptable`` must also be ``unnamed_addr``.
1220 This attribute suggests that optimization passes and code generator
1221 passes make choices that keep the code size of this function as small
1222 as possible and perform optimizations that may sacrifice runtime
1223 performance in order to minimize the size of the generated code.
1225 This attribute disables prologue / epilogue emission for the
1226 function. This can have very system-specific consequences.
1228 This indicates that the callee function at a call site is not recognized as
1229 a built-in function. LLVM will retain the original call and not replace it
1230 with equivalent code based on the semantics of the built-in function, unless
1231 the call site uses the ``builtin`` attribute. This is valid at call sites
1232 and on function declarations and definitions.
1234 This attribute indicates that calls to the function cannot be
1235 duplicated. A call to a ``noduplicate`` function may be moved
1236 within its parent function, but may not be duplicated within
1237 its parent function.
1239 A function containing a ``noduplicate`` call may still
1240 be an inlining candidate, provided that the call is not
1241 duplicated by inlining. That implies that the function has
1242 internal linkage and only has one call site, so the original
1243 call is dead after inlining.
1245 This attributes disables implicit floating point instructions.
1247 This attribute indicates that the inliner should never inline this
1248 function in any situation. This attribute may not be used together
1249 with the ``alwaysinline`` attribute.
1251 This attribute suppresses lazy symbol binding for the function. This
1252 may make calls to the function faster, at the cost of extra program
1253 startup time if the function is not called during program startup.
1255 This attribute indicates that the code generator should not use a
1256 red zone, even if the target-specific ABI normally permits it.
1258 This function attribute indicates that the function never returns
1259 normally. This produces undefined behavior at runtime if the
1260 function ever does dynamically return.
1262 This function attribute indicates that the function never raises an
1263 exception. If the function does raise an exception, its runtime
1264 behavior is undefined. However, functions marked nounwind may still
1265 trap or generate asynchronous exceptions. Exception handling schemes
1266 that are recognized by LLVM to handle asynchronous exceptions, such
1267 as SEH, will still provide their implementation defined semantics.
1269 This function attribute indicates that the function is not optimized
1270 by any optimization or code generator passes with the
1271 exception of interprocedural optimization passes.
1272 This attribute cannot be used together with the ``alwaysinline``
1273 attribute; this attribute is also incompatible
1274 with the ``minsize`` attribute and the ``optsize`` attribute.
1276 This attribute requires the ``noinline`` attribute to be specified on
1277 the function as well, so the function is never inlined into any caller.
1278 Only functions with the ``alwaysinline`` attribute are valid
1279 candidates for inlining into the body of this function.
1281 This attribute suggests that optimization passes and code generator
1282 passes make choices that keep the code size of this function low,
1283 and otherwise do optimizations specifically to reduce code size as
1284 long as they do not significantly impact runtime performance.
1286 On a function, this attribute indicates that the function computes its
1287 result (or decides to unwind an exception) based strictly on its arguments,
1288 without dereferencing any pointer arguments or otherwise accessing
1289 any mutable state (e.g. memory, control registers, etc) visible to
1290 caller functions. It does not write through any pointer arguments
1291 (including ``byval`` arguments) and never changes any state visible
1292 to callers. This means that it cannot unwind exceptions by calling
1293 the ``C++`` exception throwing methods.
1295 On an argument, this attribute indicates that the function does not
1296 dereference that pointer argument, even though it may read or write the
1297 memory that the pointer points to if accessed through other pointers.
1299 On a function, this attribute indicates that the function does not write
1300 through any pointer arguments (including ``byval`` arguments) or otherwise
1301 modify any state (e.g. memory, control registers, etc) visible to
1302 caller functions. It may dereference pointer arguments and read
1303 state that may be set in the caller. A readonly function always
1304 returns the same value (or unwinds an exception identically) when
1305 called with the same set of arguments and global state. It cannot
1306 unwind an exception by calling the ``C++`` exception throwing
1309 On an argument, this attribute indicates that the function does not write
1310 through this pointer argument, even though it may write to the memory that
1311 the pointer points to.
1313 This attribute indicates that this function can return twice. The C
1314 ``setjmp`` is an example of such a function. The compiler disables
1315 some optimizations (like tail calls) in the caller of these
1317 ``sanitize_address``
1318 This attribute indicates that AddressSanitizer checks
1319 (dynamic address safety analysis) are enabled for this function.
1321 This attribute indicates that MemorySanitizer checks (dynamic detection
1322 of accesses to uninitialized memory) are enabled for this function.
1324 This attribute indicates that ThreadSanitizer checks
1325 (dynamic thread safety analysis) are enabled for this function.
1327 This attribute indicates that the function should emit a stack
1328 smashing protector. It is in the form of a "canary" --- a random value
1329 placed on the stack before the local variables that's checked upon
1330 return from the function to see if it has been overwritten. A
1331 heuristic is used to determine if a function needs stack protectors
1332 or not. The heuristic used will enable protectors for functions with:
1334 - Character arrays larger than ``ssp-buffer-size`` (default 8).
1335 - Aggregates containing character arrays larger than ``ssp-buffer-size``.
1336 - Calls to alloca() with variable sizes or constant sizes greater than
1337 ``ssp-buffer-size``.
1339 Variables that are identified as requiring a protector will be arranged
1340 on the stack such that they are adjacent to the stack protector guard.
1342 If a function that has an ``ssp`` attribute is inlined into a
1343 function that doesn't have an ``ssp`` attribute, then the resulting
1344 function will have an ``ssp`` attribute.
1346 This attribute indicates that the function should *always* emit a
1347 stack smashing protector. This overrides the ``ssp`` function
1350 Variables that are identified as requiring a protector will be arranged
1351 on the stack such that they are adjacent to the stack protector guard.
1352 The specific layout rules are:
1354 #. Large arrays and structures containing large arrays
1355 (``>= ssp-buffer-size``) are closest to the stack protector.
1356 #. Small arrays and structures containing small arrays
1357 (``< ssp-buffer-size``) are 2nd closest to the protector.
1358 #. Variables that have had their address taken are 3rd closest to the
1361 If a function that has an ``sspreq`` attribute is inlined into a
1362 function that doesn't have an ``sspreq`` attribute or which has an
1363 ``ssp`` or ``sspstrong`` attribute, then the resulting function will have
1364 an ``sspreq`` attribute.
1366 This attribute indicates that the function should emit a stack smashing
1367 protector. This attribute causes a strong heuristic to be used when
1368 determining if a function needs stack protectors. The strong heuristic
1369 will enable protectors for functions with:
1371 - Arrays of any size and type
1372 - Aggregates containing an array of any size and type.
1373 - Calls to alloca().
1374 - Local variables that have had their address taken.
1376 Variables that are identified as requiring a protector will be arranged
1377 on the stack such that they are adjacent to the stack protector guard.
1378 The specific layout rules are:
1380 #. Large arrays and structures containing large arrays
1381 (``>= ssp-buffer-size``) are closest to the stack protector.
1382 #. Small arrays and structures containing small arrays
1383 (``< ssp-buffer-size``) are 2nd closest to the protector.
1384 #. Variables that have had their address taken are 3rd closest to the
1387 This overrides the ``ssp`` function attribute.
1389 If a function that has an ``sspstrong`` attribute is inlined into a
1390 function that doesn't have an ``sspstrong`` attribute, then the
1391 resulting function will have an ``sspstrong`` attribute.
1393 This attribute indicates that the function will delegate to some other
1394 function with a tail call. The prototype of a thunk should not be used for
1395 optimization purposes. The caller is expected to cast the thunk prototype to
1396 match the thunk target prototype.
1398 This attribute indicates that the ABI being targeted requires that
1399 an unwind table entry be produce for this function even if we can
1400 show that no exceptions passes by it. This is normally the case for
1401 the ELF x86-64 abi, but it can be disabled for some compilation
1406 Module-Level Inline Assembly
1407 ----------------------------
1409 Modules may contain "module-level inline asm" blocks, which corresponds
1410 to the GCC "file scope inline asm" blocks. These blocks are internally
1411 concatenated by LLVM and treated as a single unit, but may be separated
1412 in the ``.ll`` file if desired. The syntax is very simple:
1414 .. code-block:: llvm
1416 module asm "inline asm code goes here"
1417 module asm "more can go here"
1419 The strings can contain any character by escaping non-printable
1420 characters. The escape sequence used is simply "\\xx" where "xx" is the
1421 two digit hex code for the number.
1423 The inline asm code is simply printed to the machine code .s file when
1424 assembly code is generated.
1426 .. _langref_datalayout:
1431 A module may specify a target specific data layout string that specifies
1432 how data is to be laid out in memory. The syntax for the data layout is
1435 .. code-block:: llvm
1437 target datalayout = "layout specification"
1439 The *layout specification* consists of a list of specifications
1440 separated by the minus sign character ('-'). Each specification starts
1441 with a letter and may include other information after the letter to
1442 define some aspect of the data layout. The specifications accepted are
1446 Specifies that the target lays out data in big-endian form. That is,
1447 the bits with the most significance have the lowest address
1450 Specifies that the target lays out data in little-endian form. That
1451 is, the bits with the least significance have the lowest address
1454 Specifies the natural alignment of the stack in bits. Alignment
1455 promotion of stack variables is limited to the natural stack
1456 alignment to avoid dynamic stack realignment. The stack alignment
1457 must be a multiple of 8-bits. If omitted, the natural stack
1458 alignment defaults to "unspecified", which does not prevent any
1459 alignment promotions.
1460 ``p[n]:<size>:<abi>:<pref>``
1461 This specifies the *size* of a pointer and its ``<abi>`` and
1462 ``<pref>``\erred alignments for address space ``n``. All sizes are in
1463 bits. The address space, ``n`` is optional, and if not specified,
1464 denotes the default address space 0. The value of ``n`` must be
1465 in the range [1,2^23).
1466 ``i<size>:<abi>:<pref>``
1467 This specifies the alignment for an integer type of a given bit
1468 ``<size>``. The value of ``<size>`` must be in the range [1,2^23).
1469 ``v<size>:<abi>:<pref>``
1470 This specifies the alignment for a vector type of a given bit
1472 ``f<size>:<abi>:<pref>``
1473 This specifies the alignment for a floating point type of a given bit
1474 ``<size>``. Only values of ``<size>`` that are supported by the target
1475 will work. 32 (float) and 64 (double) are supported on all targets; 80
1476 or 128 (different flavors of long double) are also supported on some
1479 This specifies the alignment for an object of aggregate type.
1481 If present, specifies that llvm names are mangled in the output. The
1484 * ``e``: ELF mangling: Private symbols get a ``.L`` prefix.
1485 * ``m``: Mips mangling: Private symbols get a ``$`` prefix.
1486 * ``o``: Mach-O mangling: Private symbols get ``L`` prefix. Other
1487 symbols get a ``_`` prefix.
1488 * ``w``: Windows COFF prefix: Similar to Mach-O, but stdcall and fastcall
1489 functions also get a suffix based on the frame size.
1490 ``n<size1>:<size2>:<size3>...``
1491 This specifies a set of native integer widths for the target CPU in
1492 bits. For example, it might contain ``n32`` for 32-bit PowerPC,
1493 ``n32:64`` for PowerPC 64, or ``n8:16:32:64`` for X86-64. Elements of
1494 this set are considered to support most general arithmetic operations
1497 On every specification that takes a ``<abi>:<pref>``, specifying the
1498 ``<pref>`` alignment is optional. If omitted, the preceding ``:``
1499 should be omitted too and ``<pref>`` will be equal to ``<abi>``.
1501 When constructing the data layout for a given target, LLVM starts with a
1502 default set of specifications which are then (possibly) overridden by
1503 the specifications in the ``datalayout`` keyword. The default
1504 specifications are given in this list:
1506 - ``E`` - big endian
1507 - ``p:64:64:64`` - 64-bit pointers with 64-bit alignment.
1508 - ``p[n]:64:64:64`` - Other address spaces are assumed to be the
1509 same as the default address space.
1510 - ``S0`` - natural stack alignment is unspecified
1511 - ``i1:8:8`` - i1 is 8-bit (byte) aligned
1512 - ``i8:8:8`` - i8 is 8-bit (byte) aligned
1513 - ``i16:16:16`` - i16 is 16-bit aligned
1514 - ``i32:32:32`` - i32 is 32-bit aligned
1515 - ``i64:32:64`` - i64 has ABI alignment of 32-bits but preferred
1516 alignment of 64-bits
1517 - ``f16:16:16`` - half is 16-bit aligned
1518 - ``f32:32:32`` - float is 32-bit aligned
1519 - ``f64:64:64`` - double is 64-bit aligned
1520 - ``f128:128:128`` - quad is 128-bit aligned
1521 - ``v64:64:64`` - 64-bit vector is 64-bit aligned
1522 - ``v128:128:128`` - 128-bit vector is 128-bit aligned
1523 - ``a:0:64`` - aggregates are 64-bit aligned
1525 When LLVM is determining the alignment for a given type, it uses the
1528 #. If the type sought is an exact match for one of the specifications,
1529 that specification is used.
1530 #. If no match is found, and the type sought is an integer type, then
1531 the smallest integer type that is larger than the bitwidth of the
1532 sought type is used. If none of the specifications are larger than
1533 the bitwidth then the largest integer type is used. For example,
1534 given the default specifications above, the i7 type will use the
1535 alignment of i8 (next largest) while both i65 and i256 will use the
1536 alignment of i64 (largest specified).
1537 #. If no match is found, and the type sought is a vector type, then the
1538 largest vector type that is smaller than the sought vector type will
1539 be used as a fall back. This happens because <128 x double> can be
1540 implemented in terms of 64 <2 x double>, for example.
1542 The function of the data layout string may not be what you expect.
1543 Notably, this is not a specification from the frontend of what alignment
1544 the code generator should use.
1546 Instead, if specified, the target data layout is required to match what
1547 the ultimate *code generator* expects. This string is used by the
1548 mid-level optimizers to improve code, and this only works if it matches
1549 what the ultimate code generator uses. There is no way to generate IR
1550 that does not embed this target-specific detail into the IR. If you
1551 don't specify the string, the default specifications will be used to
1552 generate a Data Layout and the optimization phases will operate
1553 accordingly and introduce target specificity into the IR with respect to
1554 these default specifications.
1561 A module may specify a target triple string that describes the target
1562 host. The syntax for the target triple is simply:
1564 .. code-block:: llvm
1566 target triple = "x86_64-apple-macosx10.7.0"
1568 The *target triple* string consists of a series of identifiers delimited
1569 by the minus sign character ('-'). The canonical forms are:
1573 ARCHITECTURE-VENDOR-OPERATING_SYSTEM
1574 ARCHITECTURE-VENDOR-OPERATING_SYSTEM-ENVIRONMENT
1576 This information is passed along to the backend so that it generates
1577 code for the proper architecture. It's possible to override this on the
1578 command line with the ``-mtriple`` command line option.
1580 .. _pointeraliasing:
1582 Pointer Aliasing Rules
1583 ----------------------
1585 Any memory access must be done through a pointer value associated with
1586 an address range of the memory access, otherwise the behavior is
1587 undefined. Pointer values are associated with address ranges according
1588 to the following rules:
1590 - A pointer value is associated with the addresses associated with any
1591 value it is *based* on.
1592 - An address of a global variable is associated with the address range
1593 of the variable's storage.
1594 - The result value of an allocation instruction is associated with the
1595 address range of the allocated storage.
1596 - A null pointer in the default address-space is associated with no
1598 - An integer constant other than zero or a pointer value returned from
1599 a function not defined within LLVM may be associated with address
1600 ranges allocated through mechanisms other than those provided by
1601 LLVM. Such ranges shall not overlap with any ranges of addresses
1602 allocated by mechanisms provided by LLVM.
1604 A pointer value is *based* on another pointer value according to the
1607 - A pointer value formed from a ``getelementptr`` operation is *based*
1608 on the first value operand of the ``getelementptr``.
1609 - The result value of a ``bitcast`` is *based* on the operand of the
1611 - A pointer value formed by an ``inttoptr`` is *based* on all pointer
1612 values that contribute (directly or indirectly) to the computation of
1613 the pointer's value.
1614 - The "*based* on" relationship is transitive.
1616 Note that this definition of *"based"* is intentionally similar to the
1617 definition of *"based"* in C99, though it is slightly weaker.
1619 LLVM IR does not associate types with memory. The result type of a
1620 ``load`` merely indicates the size and alignment of the memory from
1621 which to load, as well as the interpretation of the value. The first
1622 operand type of a ``store`` similarly only indicates the size and
1623 alignment of the store.
1625 Consequently, type-based alias analysis, aka TBAA, aka
1626 ``-fstrict-aliasing``, is not applicable to general unadorned LLVM IR.
1627 :ref:`Metadata <metadata>` may be used to encode additional information
1628 which specialized optimization passes may use to implement type-based
1633 Volatile Memory Accesses
1634 ------------------------
1636 Certain memory accesses, such as :ref:`load <i_load>`'s,
1637 :ref:`store <i_store>`'s, and :ref:`llvm.memcpy <int_memcpy>`'s may be
1638 marked ``volatile``. The optimizers must not change the number of
1639 volatile operations or change their order of execution relative to other
1640 volatile operations. The optimizers *may* change the order of volatile
1641 operations relative to non-volatile operations. This is not Java's
1642 "volatile" and has no cross-thread synchronization behavior.
1644 IR-level volatile loads and stores cannot safely be optimized into
1645 llvm.memcpy or llvm.memmove intrinsics even when those intrinsics are
1646 flagged volatile. Likewise, the backend should never split or merge
1647 target-legal volatile load/store instructions.
1649 .. admonition:: Rationale
1651 Platforms may rely on volatile loads and stores of natively supported
1652 data width to be executed as single instruction. For example, in C
1653 this holds for an l-value of volatile primitive type with native
1654 hardware support, but not necessarily for aggregate types. The
1655 frontend upholds these expectations, which are intentionally
1656 unspecified in the IR. The rules above ensure that IR transformation
1657 do not violate the frontend's contract with the language.
1661 Memory Model for Concurrent Operations
1662 --------------------------------------
1664 The LLVM IR does not define any way to start parallel threads of
1665 execution or to register signal handlers. Nonetheless, there are
1666 platform-specific ways to create them, and we define LLVM IR's behavior
1667 in their presence. This model is inspired by the C++0x memory model.
1669 For a more informal introduction to this model, see the :doc:`Atomics`.
1671 We define a *happens-before* partial order as the least partial order
1674 - Is a superset of single-thread program order, and
1675 - When a *synchronizes-with* ``b``, includes an edge from ``a`` to
1676 ``b``. *Synchronizes-with* pairs are introduced by platform-specific
1677 techniques, like pthread locks, thread creation, thread joining,
1678 etc., and by atomic instructions. (See also :ref:`Atomic Memory Ordering
1679 Constraints <ordering>`).
1681 Note that program order does not introduce *happens-before* edges
1682 between a thread and signals executing inside that thread.
1684 Every (defined) read operation (load instructions, memcpy, atomic
1685 loads/read-modify-writes, etc.) R reads a series of bytes written by
1686 (defined) write operations (store instructions, atomic
1687 stores/read-modify-writes, memcpy, etc.). For the purposes of this
1688 section, initialized globals are considered to have a write of the
1689 initializer which is atomic and happens before any other read or write
1690 of the memory in question. For each byte of a read R, R\ :sub:`byte`
1691 may see any write to the same byte, except:
1693 - If write\ :sub:`1` happens before write\ :sub:`2`, and
1694 write\ :sub:`2` happens before R\ :sub:`byte`, then
1695 R\ :sub:`byte` does not see write\ :sub:`1`.
1696 - If R\ :sub:`byte` happens before write\ :sub:`3`, then
1697 R\ :sub:`byte` does not see write\ :sub:`3`.
1699 Given that definition, R\ :sub:`byte` is defined as follows:
1701 - If R is volatile, the result is target-dependent. (Volatile is
1702 supposed to give guarantees which can support ``sig_atomic_t`` in
1703 C/C++, and may be used for accesses to addresses that do not behave
1704 like normal memory. It does not generally provide cross-thread
1706 - Otherwise, if there is no write to the same byte that happens before
1707 R\ :sub:`byte`, R\ :sub:`byte` returns ``undef`` for that byte.
1708 - Otherwise, if R\ :sub:`byte` may see exactly one write,
1709 R\ :sub:`byte` returns the value written by that write.
1710 - Otherwise, if R is atomic, and all the writes R\ :sub:`byte` may
1711 see are atomic, it chooses one of the values written. See the :ref:`Atomic
1712 Memory Ordering Constraints <ordering>` section for additional
1713 constraints on how the choice is made.
1714 - Otherwise R\ :sub:`byte` returns ``undef``.
1716 R returns the value composed of the series of bytes it read. This
1717 implies that some bytes within the value may be ``undef`` **without**
1718 the entire value being ``undef``. Note that this only defines the
1719 semantics of the operation; it doesn't mean that targets will emit more
1720 than one instruction to read the series of bytes.
1722 Note that in cases where none of the atomic intrinsics are used, this
1723 model places only one restriction on IR transformations on top of what
1724 is required for single-threaded execution: introducing a store to a byte
1725 which might not otherwise be stored is not allowed in general.
1726 (Specifically, in the case where another thread might write to and read
1727 from an address, introducing a store can change a load that may see
1728 exactly one write into a load that may see multiple writes.)
1732 Atomic Memory Ordering Constraints
1733 ----------------------------------
1735 Atomic instructions (:ref:`cmpxchg <i_cmpxchg>`,
1736 :ref:`atomicrmw <i_atomicrmw>`, :ref:`fence <i_fence>`,
1737 :ref:`atomic load <i_load>`, and :ref:`atomic store <i_store>`) take
1738 ordering parameters that determine which other atomic instructions on
1739 the same address they *synchronize with*. These semantics are borrowed
1740 from Java and C++0x, but are somewhat more colloquial. If these
1741 descriptions aren't precise enough, check those specs (see spec
1742 references in the :doc:`atomics guide <Atomics>`).
1743 :ref:`fence <i_fence>` instructions treat these orderings somewhat
1744 differently since they don't take an address. See that instruction's
1745 documentation for details.
1747 For a simpler introduction to the ordering constraints, see the
1751 The set of values that can be read is governed by the happens-before
1752 partial order. A value cannot be read unless some operation wrote
1753 it. This is intended to provide a guarantee strong enough to model
1754 Java's non-volatile shared variables. This ordering cannot be
1755 specified for read-modify-write operations; it is not strong enough
1756 to make them atomic in any interesting way.
1758 In addition to the guarantees of ``unordered``, there is a single
1759 total order for modifications by ``monotonic`` operations on each
1760 address. All modification orders must be compatible with the
1761 happens-before order. There is no guarantee that the modification
1762 orders can be combined to a global total order for the whole program
1763 (and this often will not be possible). The read in an atomic
1764 read-modify-write operation (:ref:`cmpxchg <i_cmpxchg>` and
1765 :ref:`atomicrmw <i_atomicrmw>`) reads the value in the modification
1766 order immediately before the value it writes. If one atomic read
1767 happens before another atomic read of the same address, the later
1768 read must see the same value or a later value in the address's
1769 modification order. This disallows reordering of ``monotonic`` (or
1770 stronger) operations on the same address. If an address is written
1771 ``monotonic``-ally by one thread, and other threads ``monotonic``-ally
1772 read that address repeatedly, the other threads must eventually see
1773 the write. This corresponds to the C++0x/C1x
1774 ``memory_order_relaxed``.
1776 In addition to the guarantees of ``monotonic``, a
1777 *synchronizes-with* edge may be formed with a ``release`` operation.
1778 This is intended to model C++'s ``memory_order_acquire``.
1780 In addition to the guarantees of ``monotonic``, if this operation
1781 writes a value which is subsequently read by an ``acquire``
1782 operation, it *synchronizes-with* that operation. (This isn't a
1783 complete description; see the C++0x definition of a release
1784 sequence.) This corresponds to the C++0x/C1x
1785 ``memory_order_release``.
1786 ``acq_rel`` (acquire+release)
1787 Acts as both an ``acquire`` and ``release`` operation on its
1788 address. This corresponds to the C++0x/C1x ``memory_order_acq_rel``.
1789 ``seq_cst`` (sequentially consistent)
1790 In addition to the guarantees of ``acq_rel`` (``acquire`` for an
1791 operation that only reads, ``release`` for an operation that only
1792 writes), there is a global total order on all
1793 sequentially-consistent operations on all addresses, which is
1794 consistent with the *happens-before* partial order and with the
1795 modification orders of all the affected addresses. Each
1796 sequentially-consistent read sees the last preceding write to the
1797 same address in this global order. This corresponds to the C++0x/C1x
1798 ``memory_order_seq_cst`` and Java volatile.
1802 If an atomic operation is marked ``singlethread``, it only *synchronizes
1803 with* or participates in modification and seq\_cst total orderings with
1804 other operations running in the same thread (for example, in signal
1812 LLVM IR floating-point binary ops (:ref:`fadd <i_fadd>`,
1813 :ref:`fsub <i_fsub>`, :ref:`fmul <i_fmul>`, :ref:`fdiv <i_fdiv>`,
1814 :ref:`frem <i_frem>`) have the following flags that can be set to enable
1815 otherwise unsafe floating point operations
1818 No NaNs - Allow optimizations to assume the arguments and result are not
1819 NaN. Such optimizations are required to retain defined behavior over
1820 NaNs, but the value of the result is undefined.
1823 No Infs - Allow optimizations to assume the arguments and result are not
1824 +/-Inf. Such optimizations are required to retain defined behavior over
1825 +/-Inf, but the value of the result is undefined.
1828 No Signed Zeros - Allow optimizations to treat the sign of a zero
1829 argument or result as insignificant.
1832 Allow Reciprocal - Allow optimizations to use the reciprocal of an
1833 argument rather than perform division.
1836 Fast - Allow algebraically equivalent transformations that may
1837 dramatically change results in floating point (e.g. reassociate). This
1838 flag implies all the others.
1842 Use-list Order Directives
1843 -------------------------
1845 Use-list directives encode the in-memory order of each use-list, allowing the
1846 order to be recreated. ``<order-indexes>`` is a comma-separated list of
1847 indexes that are assigned to the referenced value's uses. The referenced
1848 value's use-list is immediately sorted by these indexes.
1850 Use-list directives may appear at function scope or global scope. They are not
1851 instructions, and have no effect on the semantics of the IR. When they're at
1852 function scope, they must appear after the terminator of the final basic block.
1854 If basic blocks have their address taken via ``blockaddress()`` expressions,
1855 ``uselistorder_bb`` can be used to reorder their use-lists from outside their
1862 uselistorder <ty> <value>, { <order-indexes> }
1863 uselistorder_bb @function, %block { <order-indexes> }
1869 define void @foo(i32 %arg1, i32 %arg2) {
1871 ; ... instructions ...
1873 ; ... instructions ...
1875 ; At function scope.
1876 uselistorder i32 %arg1, { 1, 0, 2 }
1877 uselistorder label %bb, { 1, 0 }
1881 uselistorder i32* @global, { 1, 2, 0 }
1882 uselistorder i32 7, { 1, 0 }
1883 uselistorder i32 (i32) @bar, { 1, 0 }
1884 uselistorder_bb @foo, %bb, { 5, 1, 3, 2, 0, 4 }
1891 The LLVM type system is one of the most important features of the
1892 intermediate representation. Being typed enables a number of
1893 optimizations to be performed on the intermediate representation
1894 directly, without having to do extra analyses on the side before the
1895 transformation. A strong type system makes it easier to read the
1896 generated code and enables novel analyses and transformations that are
1897 not feasible to perform on normal three address code representations.
1907 The void type does not represent any value and has no size.
1925 The function type can be thought of as a function signature. It consists of a
1926 return type and a list of formal parameter types. The return type of a function
1927 type is a void type or first class type --- except for :ref:`label <t_label>`
1928 and :ref:`metadata <t_metadata>` types.
1934 <returntype> (<parameter list>)
1936 ...where '``<parameter list>``' is a comma-separated list of type
1937 specifiers. Optionally, the parameter list may include a type ``...``, which
1938 indicates that the function takes a variable number of arguments. Variable
1939 argument functions can access their arguments with the :ref:`variable argument
1940 handling intrinsic <int_varargs>` functions. '``<returntype>``' is any type
1941 except :ref:`label <t_label>` and :ref:`metadata <t_metadata>`.
1945 +---------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------+
1946 | ``i32 (i32)`` | function taking an ``i32``, returning an ``i32`` |
1947 +---------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------+
1948 | ``float (i16, i32 *) *`` | :ref:`Pointer <t_pointer>` to a function that takes an ``i16`` and a :ref:`pointer <t_pointer>` to ``i32``, returning ``float``. |
1949 +---------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------+
1950 | ``i32 (i8*, ...)`` | A vararg function that takes at least one :ref:`pointer <t_pointer>` to ``i8`` (char in C), which returns an integer. This is the signature for ``printf`` in LLVM. |
1951 +---------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------+
1952 | ``{i32, i32} (i32)`` | A function taking an ``i32``, returning a :ref:`structure <t_struct>` containing two ``i32`` values |
1953 +---------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------+
1960 The :ref:`first class <t_firstclass>` types are perhaps the most important.
1961 Values of these types are the only ones which can be produced by
1969 These are the types that are valid in registers from CodeGen's perspective.
1978 The integer type is a very simple type that simply specifies an
1979 arbitrary bit width for the integer type desired. Any bit width from 1
1980 bit to 2\ :sup:`23`\ -1 (about 8 million) can be specified.
1988 The number of bits the integer will occupy is specified by the ``N``
1994 +----------------+------------------------------------------------+
1995 | ``i1`` | a single-bit integer. |
1996 +----------------+------------------------------------------------+
1997 | ``i32`` | a 32-bit integer. |
1998 +----------------+------------------------------------------------+
1999 | ``i1942652`` | a really big integer of over 1 million bits. |
2000 +----------------+------------------------------------------------+
2004 Floating Point Types
2005 """"""""""""""""""""
2014 - 16-bit floating point value
2017 - 32-bit floating point value
2020 - 64-bit floating point value
2023 - 128-bit floating point value (112-bit mantissa)
2026 - 80-bit floating point value (X87)
2029 - 128-bit floating point value (two 64-bits)
2036 The x86_mmx type represents a value held in an MMX register on an x86
2037 machine. The operations allowed on it are quite limited: parameters and
2038 return values, load and store, and bitcast. User-specified MMX
2039 instructions are represented as intrinsic or asm calls with arguments
2040 and/or results of this type. There are no arrays, vectors or constants
2057 The pointer type is used to specify memory locations. Pointers are
2058 commonly used to reference objects in memory.
2060 Pointer types may have an optional address space attribute defining the
2061 numbered address space where the pointed-to object resides. The default
2062 address space is number zero. The semantics of non-zero address spaces
2063 are target-specific.
2065 Note that LLVM does not permit pointers to void (``void*``) nor does it
2066 permit pointers to labels (``label*``). Use ``i8*`` instead.
2076 +-------------------------+--------------------------------------------------------------------------------------------------------------+
2077 | ``[4 x i32]*`` | A :ref:`pointer <t_pointer>` to :ref:`array <t_array>` of four ``i32`` values. |
2078 +-------------------------+--------------------------------------------------------------------------------------------------------------+
2079 | ``i32 (i32*) *`` | A :ref:`pointer <t_pointer>` to a :ref:`function <t_function>` that takes an ``i32*``, returning an ``i32``. |
2080 +-------------------------+--------------------------------------------------------------------------------------------------------------+
2081 | ``i32 addrspace(5)*`` | A :ref:`pointer <t_pointer>` to an ``i32`` value that resides in address space #5. |
2082 +-------------------------+--------------------------------------------------------------------------------------------------------------+
2091 A vector type is a simple derived type that represents a vector of
2092 elements. Vector types are used when multiple primitive data are
2093 operated in parallel using a single instruction (SIMD). A vector type
2094 requires a size (number of elements) and an underlying primitive data
2095 type. Vector types are considered :ref:`first class <t_firstclass>`.
2101 < <# elements> x <elementtype> >
2103 The number of elements is a constant integer value larger than 0;
2104 elementtype may be any integer, floating point or pointer type. Vectors
2105 of size zero are not allowed.
2109 +-------------------+--------------------------------------------------+
2110 | ``<4 x i32>`` | Vector of 4 32-bit integer values. |
2111 +-------------------+--------------------------------------------------+
2112 | ``<8 x float>`` | Vector of 8 32-bit floating-point values. |
2113 +-------------------+--------------------------------------------------+
2114 | ``<2 x i64>`` | Vector of 2 64-bit integer values. |
2115 +-------------------+--------------------------------------------------+
2116 | ``<4 x i64*>`` | Vector of 4 pointers to 64-bit integer values. |
2117 +-------------------+--------------------------------------------------+
2126 The label type represents code labels.
2141 The metadata type represents embedded metadata. No derived types may be
2142 created from metadata except for :ref:`function <t_function>` arguments.
2155 Aggregate Types are a subset of derived types that can contain multiple
2156 member types. :ref:`Arrays <t_array>` and :ref:`structs <t_struct>` are
2157 aggregate types. :ref:`Vectors <t_vector>` are not considered to be
2167 The array type is a very simple derived type that arranges elements
2168 sequentially in memory. The array type requires a size (number of
2169 elements) and an underlying data type.
2175 [<# elements> x <elementtype>]
2177 The number of elements is a constant integer value; ``elementtype`` may
2178 be any type with a size.
2182 +------------------+--------------------------------------+
2183 | ``[40 x i32]`` | Array of 40 32-bit integer values. |
2184 +------------------+--------------------------------------+
2185 | ``[41 x i32]`` | Array of 41 32-bit integer values. |
2186 +------------------+--------------------------------------+
2187 | ``[4 x i8]`` | Array of 4 8-bit integer values. |
2188 +------------------+--------------------------------------+
2190 Here are some examples of multidimensional arrays:
2192 +-----------------------------+----------------------------------------------------------+
2193 | ``[3 x [4 x i32]]`` | 3x4 array of 32-bit integer values. |
2194 +-----------------------------+----------------------------------------------------------+
2195 | ``[12 x [10 x float]]`` | 12x10 array of single precision floating point values. |
2196 +-----------------------------+----------------------------------------------------------+
2197 | ``[2 x [3 x [4 x i16]]]`` | 2x3x4 array of 16-bit integer values. |
2198 +-----------------------------+----------------------------------------------------------+
2200 There is no restriction on indexing beyond the end of the array implied
2201 by a static type (though there are restrictions on indexing beyond the
2202 bounds of an allocated object in some cases). This means that
2203 single-dimension 'variable sized array' addressing can be implemented in
2204 LLVM with a zero length array type. An implementation of 'pascal style
2205 arrays' in LLVM could use the type "``{ i32, [0 x float]}``", for
2215 The structure type is used to represent a collection of data members
2216 together in memory. The elements of a structure may be any type that has
2219 Structures in memory are accessed using '``load``' and '``store``' by
2220 getting a pointer to a field with the '``getelementptr``' instruction.
2221 Structures in registers are accessed using the '``extractvalue``' and
2222 '``insertvalue``' instructions.
2224 Structures may optionally be "packed" structures, which indicate that
2225 the alignment of the struct is one byte, and that there is no padding
2226 between the elements. In non-packed structs, padding between field types
2227 is inserted as defined by the DataLayout string in the module, which is
2228 required to match what the underlying code generator expects.
2230 Structures can either be "literal" or "identified". A literal structure
2231 is defined inline with other types (e.g. ``{i32, i32}*``) whereas
2232 identified types are always defined at the top level with a name.
2233 Literal types are uniqued by their contents and can never be recursive
2234 or opaque since there is no way to write one. Identified types can be
2235 recursive, can be opaqued, and are never uniqued.
2241 %T1 = type { <type list> } ; Identified normal struct type
2242 %T2 = type <{ <type list> }> ; Identified packed struct type
2246 +------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+
2247 | ``{ i32, i32, i32 }`` | A triple of three ``i32`` values |
2248 +------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+
2249 | ``{ float, i32 (i32) * }`` | A pair, where the first element is a ``float`` and the second element is a :ref:`pointer <t_pointer>` to a :ref:`function <t_function>` that takes an ``i32``, returning an ``i32``. |
2250 +------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+
2251 | ``<{ i8, i32 }>`` | A packed struct known to be 5 bytes in size. |
2252 +------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+
2256 Opaque Structure Types
2257 """"""""""""""""""""""
2261 Opaque structure types are used to represent named structure types that
2262 do not have a body specified. This corresponds (for example) to the C
2263 notion of a forward declared structure.
2274 +--------------+-------------------+
2275 | ``opaque`` | An opaque type. |
2276 +--------------+-------------------+
2283 LLVM has several different basic types of constants. This section
2284 describes them all and their syntax.
2289 **Boolean constants**
2290 The two strings '``true``' and '``false``' are both valid constants
2292 **Integer constants**
2293 Standard integers (such as '4') are constants of the
2294 :ref:`integer <t_integer>` type. Negative numbers may be used with
2296 **Floating point constants**
2297 Floating point constants use standard decimal notation (e.g.
2298 123.421), exponential notation (e.g. 1.23421e+2), or a more precise
2299 hexadecimal notation (see below). The assembler requires the exact
2300 decimal value of a floating-point constant. For example, the
2301 assembler accepts 1.25 but rejects 1.3 because 1.3 is a repeating
2302 decimal in binary. Floating point constants must have a :ref:`floating
2303 point <t_floating>` type.
2304 **Null pointer constants**
2305 The identifier '``null``' is recognized as a null pointer constant
2306 and must be of :ref:`pointer type <t_pointer>`.
2308 The one non-intuitive notation for constants is the hexadecimal form of
2309 floating point constants. For example, the form
2310 '``double 0x432ff973cafa8000``' is equivalent to (but harder to read
2311 than) '``double 4.5e+15``'. The only time hexadecimal floating point
2312 constants are required (and the only time that they are generated by the
2313 disassembler) is when a floating point constant must be emitted but it
2314 cannot be represented as a decimal floating point number in a reasonable
2315 number of digits. For example, NaN's, infinities, and other special
2316 values are represented in their IEEE hexadecimal format so that assembly
2317 and disassembly do not cause any bits to change in the constants.
2319 When using the hexadecimal form, constants of types half, float, and
2320 double are represented using the 16-digit form shown above (which
2321 matches the IEEE754 representation for double); half and float values
2322 must, however, be exactly representable as IEEE 754 half and single
2323 precision, respectively. Hexadecimal format is always used for long
2324 double, and there are three forms of long double. The 80-bit format used
2325 by x86 is represented as ``0xK`` followed by 20 hexadecimal digits. The
2326 128-bit format used by PowerPC (two adjacent doubles) is represented by
2327 ``0xM`` followed by 32 hexadecimal digits. The IEEE 128-bit format is
2328 represented by ``0xL`` followed by 32 hexadecimal digits. Long doubles
2329 will only work if they match the long double format on your target.
2330 The IEEE 16-bit format (half precision) is represented by ``0xH``
2331 followed by 4 hexadecimal digits. All hexadecimal formats are big-endian
2332 (sign bit at the left).
2334 There are no constants of type x86_mmx.
2336 .. _complexconstants:
2341 Complex constants are a (potentially recursive) combination of simple
2342 constants and smaller complex constants.
2344 **Structure constants**
2345 Structure constants are represented with notation similar to
2346 structure type definitions (a comma separated list of elements,
2347 surrounded by braces (``{}``)). For example:
2348 "``{ i32 4, float 17.0, i32* @G }``", where "``@G``" is declared as
2349 "``@G = external global i32``". Structure constants must have
2350 :ref:`structure type <t_struct>`, and the number and types of elements
2351 must match those specified by the type.
2353 Array constants are represented with notation similar to array type
2354 definitions (a comma separated list of elements, surrounded by
2355 square brackets (``[]``)). For example:
2356 "``[ i32 42, i32 11, i32 74 ]``". Array constants must have
2357 :ref:`array type <t_array>`, and the number and types of elements must
2358 match those specified by the type. As a special case, character array
2359 constants may also be represented as a double-quoted string using the ``c``
2360 prefix. For example: "``c"Hello World\0A\00"``".
2361 **Vector constants**
2362 Vector constants are represented with notation similar to vector
2363 type definitions (a comma separated list of elements, surrounded by
2364 less-than/greater-than's (``<>``)). For example:
2365 "``< i32 42, i32 11, i32 74, i32 100 >``". Vector constants
2366 must have :ref:`vector type <t_vector>`, and the number and types of
2367 elements must match those specified by the type.
2368 **Zero initialization**
2369 The string '``zeroinitializer``' can be used to zero initialize a
2370 value to zero of *any* type, including scalar and
2371 :ref:`aggregate <t_aggregate>` types. This is often used to avoid
2372 having to print large zero initializers (e.g. for large arrays) and
2373 is always exactly equivalent to using explicit zero initializers.
2375 A metadata node is a constant tuple without types. For example:
2376 "``!{!0, !{!2, !0}, !"test"}``". Metadata can reference constant values,
2377 for example: "``!{!0, i32 0, i8* @global, i64 (i64)* @function, !"str"}``".
2378 Unlike other typed constants that are meant to be interpreted as part of
2379 the instruction stream, metadata is a place to attach additional
2380 information such as debug info.
2382 Global Variable and Function Addresses
2383 --------------------------------------
2385 The addresses of :ref:`global variables <globalvars>` and
2386 :ref:`functions <functionstructure>` are always implicitly valid
2387 (link-time) constants. These constants are explicitly referenced when
2388 the :ref:`identifier for the global <identifiers>` is used and always have
2389 :ref:`pointer <t_pointer>` type. For example, the following is a legal LLVM
2392 .. code-block:: llvm
2396 @Z = global [2 x i32*] [ i32* @X, i32* @Y ]
2403 The string '``undef``' can be used anywhere a constant is expected, and
2404 indicates that the user of the value may receive an unspecified
2405 bit-pattern. Undefined values may be of any type (other than '``label``'
2406 or '``void``') and be used anywhere a constant is permitted.
2408 Undefined values are useful because they indicate to the compiler that
2409 the program is well defined no matter what value is used. This gives the
2410 compiler more freedom to optimize. Here are some examples of
2411 (potentially surprising) transformations that are valid (in pseudo IR):
2413 .. code-block:: llvm
2423 This is safe because all of the output bits are affected by the undef
2424 bits. Any output bit can have a zero or one depending on the input bits.
2426 .. code-block:: llvm
2437 These logical operations have bits that are not always affected by the
2438 input. For example, if ``%X`` has a zero bit, then the output of the
2439 '``and``' operation will always be a zero for that bit, no matter what
2440 the corresponding bit from the '``undef``' is. As such, it is unsafe to
2441 optimize or assume that the result of the '``and``' is '``undef``'.
2442 However, it is safe to assume that all bits of the '``undef``' could be
2443 0, and optimize the '``and``' to 0. Likewise, it is safe to assume that
2444 all the bits of the '``undef``' operand to the '``or``' could be set,
2445 allowing the '``or``' to be folded to -1.
2447 .. code-block:: llvm
2449 %A = select undef, %X, %Y
2450 %B = select undef, 42, %Y
2451 %C = select %X, %Y, undef
2461 This set of examples shows that undefined '``select``' (and conditional
2462 branch) conditions can go *either way*, but they have to come from one
2463 of the two operands. In the ``%A`` example, if ``%X`` and ``%Y`` were
2464 both known to have a clear low bit, then ``%A`` would have to have a
2465 cleared low bit. However, in the ``%C`` example, the optimizer is
2466 allowed to assume that the '``undef``' operand could be the same as
2467 ``%Y``, allowing the whole '``select``' to be eliminated.
2469 .. code-block:: llvm
2471 %A = xor undef, undef
2488 This example points out that two '``undef``' operands are not
2489 necessarily the same. This can be surprising to people (and also matches
2490 C semantics) where they assume that "``X^X``" is always zero, even if
2491 ``X`` is undefined. This isn't true for a number of reasons, but the
2492 short answer is that an '``undef``' "variable" can arbitrarily change
2493 its value over its "live range". This is true because the variable
2494 doesn't actually *have a live range*. Instead, the value is logically
2495 read from arbitrary registers that happen to be around when needed, so
2496 the value is not necessarily consistent over time. In fact, ``%A`` and
2497 ``%C`` need to have the same semantics or the core LLVM "replace all
2498 uses with" concept would not hold.
2500 .. code-block:: llvm
2508 These examples show the crucial difference between an *undefined value*
2509 and *undefined behavior*. An undefined value (like '``undef``') is
2510 allowed to have an arbitrary bit-pattern. This means that the ``%A``
2511 operation can be constant folded to '``undef``', because the '``undef``'
2512 could be an SNaN, and ``fdiv`` is not (currently) defined on SNaN's.
2513 However, in the second example, we can make a more aggressive
2514 assumption: because the ``undef`` is allowed to be an arbitrary value,
2515 we are allowed to assume that it could be zero. Since a divide by zero
2516 has *undefined behavior*, we are allowed to assume that the operation
2517 does not execute at all. This allows us to delete the divide and all
2518 code after it. Because the undefined operation "can't happen", the
2519 optimizer can assume that it occurs in dead code.
2521 .. code-block:: llvm
2523 a: store undef -> %X
2524 b: store %X -> undef
2529 These examples reiterate the ``fdiv`` example: a store *of* an undefined
2530 value can be assumed to not have any effect; we can assume that the
2531 value is overwritten with bits that happen to match what was already
2532 there. However, a store *to* an undefined location could clobber
2533 arbitrary memory, therefore, it has undefined behavior.
2540 Poison values are similar to :ref:`undef values <undefvalues>`, however
2541 they also represent the fact that an instruction or constant expression
2542 that cannot evoke side effects has nevertheless detected a condition
2543 that results in undefined behavior.
2545 There is currently no way of representing a poison value in the IR; they
2546 only exist when produced by operations such as :ref:`add <i_add>` with
2549 Poison value behavior is defined in terms of value *dependence*:
2551 - Values other than :ref:`phi <i_phi>` nodes depend on their operands.
2552 - :ref:`Phi <i_phi>` nodes depend on the operand corresponding to
2553 their dynamic predecessor basic block.
2554 - Function arguments depend on the corresponding actual argument values
2555 in the dynamic callers of their functions.
2556 - :ref:`Call <i_call>` instructions depend on the :ref:`ret <i_ret>`
2557 instructions that dynamically transfer control back to them.
2558 - :ref:`Invoke <i_invoke>` instructions depend on the
2559 :ref:`ret <i_ret>`, :ref:`resume <i_resume>`, or exception-throwing
2560 call instructions that dynamically transfer control back to them.
2561 - Non-volatile loads and stores depend on the most recent stores to all
2562 of the referenced memory addresses, following the order in the IR
2563 (including loads and stores implied by intrinsics such as
2564 :ref:`@llvm.memcpy <int_memcpy>`.)
2565 - An instruction with externally visible side effects depends on the
2566 most recent preceding instruction with externally visible side
2567 effects, following the order in the IR. (This includes :ref:`volatile
2568 operations <volatile>`.)
2569 - An instruction *control-depends* on a :ref:`terminator
2570 instruction <terminators>` if the terminator instruction has
2571 multiple successors and the instruction is always executed when
2572 control transfers to one of the successors, and may not be executed
2573 when control is transferred to another.
2574 - Additionally, an instruction also *control-depends* on a terminator
2575 instruction if the set of instructions it otherwise depends on would
2576 be different if the terminator had transferred control to a different
2578 - Dependence is transitive.
2580 Poison values have the same behavior as :ref:`undef values <undefvalues>`,
2581 with the additional effect that any instruction that has a *dependence*
2582 on a poison value has undefined behavior.
2584 Here are some examples:
2586 .. code-block:: llvm
2589 %poison = sub nuw i32 0, 1 ; Results in a poison value.
2590 %still_poison = and i32 %poison, 0 ; 0, but also poison.
2591 %poison_yet_again = getelementptr i32, i32* @h, i32 %still_poison
2592 store i32 0, i32* %poison_yet_again ; memory at @h[0] is poisoned
2594 store i32 %poison, i32* @g ; Poison value stored to memory.
2595 %poison2 = load i32, i32* @g ; Poison value loaded back from memory.
2597 store volatile i32 %poison, i32* @g ; External observation; undefined behavior.
2599 %narrowaddr = bitcast i32* @g to i16*
2600 %wideaddr = bitcast i32* @g to i64*
2601 %poison3 = load i16, i16* %narrowaddr ; Returns a poison value.
2602 %poison4 = load i64, i64* %wideaddr ; Returns a poison value.
2604 %cmp = icmp slt i32 %poison, 0 ; Returns a poison value.
2605 br i1 %cmp, label %true, label %end ; Branch to either destination.
2608 store volatile i32 0, i32* @g ; This is control-dependent on %cmp, so
2609 ; it has undefined behavior.
2613 %p = phi i32 [ 0, %entry ], [ 1, %true ]
2614 ; Both edges into this PHI are
2615 ; control-dependent on %cmp, so this
2616 ; always results in a poison value.
2618 store volatile i32 0, i32* @g ; This would depend on the store in %true
2619 ; if %cmp is true, or the store in %entry
2620 ; otherwise, so this is undefined behavior.
2622 br i1 %cmp, label %second_true, label %second_end
2623 ; The same branch again, but this time the
2624 ; true block doesn't have side effects.
2631 store volatile i32 0, i32* @g ; This time, the instruction always depends
2632 ; on the store in %end. Also, it is
2633 ; control-equivalent to %end, so this is
2634 ; well-defined (ignoring earlier undefined
2635 ; behavior in this example).
2639 Addresses of Basic Blocks
2640 -------------------------
2642 ``blockaddress(@function, %block)``
2644 The '``blockaddress``' constant computes the address of the specified
2645 basic block in the specified function, and always has an ``i8*`` type.
2646 Taking the address of the entry block is illegal.
2648 This value only has defined behavior when used as an operand to the
2649 ':ref:`indirectbr <i_indirectbr>`' instruction, or for comparisons
2650 against null. Pointer equality tests between labels addresses results in
2651 undefined behavior --- though, again, comparison against null is ok, and
2652 no label is equal to the null pointer. This may be passed around as an
2653 opaque pointer sized value as long as the bits are not inspected. This
2654 allows ``ptrtoint`` and arithmetic to be performed on these values so
2655 long as the original value is reconstituted before the ``indirectbr``
2658 Finally, some targets may provide defined semantics when using the value
2659 as the operand to an inline assembly, but that is target specific.
2663 Constant Expressions
2664 --------------------
2666 Constant expressions are used to allow expressions involving other
2667 constants to be used as constants. Constant expressions may be of any
2668 :ref:`first class <t_firstclass>` type and may involve any LLVM operation
2669 that does not have side effects (e.g. load and call are not supported).
2670 The following is the syntax for constant expressions:
2672 ``trunc (CST to TYPE)``
2673 Truncate a constant to another type. The bit size of CST must be
2674 larger than the bit size of TYPE. Both types must be integers.
2675 ``zext (CST to TYPE)``
2676 Zero extend a constant to another type. The bit size of CST must be
2677 smaller than the bit size of TYPE. Both types must be integers.
2678 ``sext (CST to TYPE)``
2679 Sign extend a constant to another type. The bit size of CST must be
2680 smaller than the bit size of TYPE. Both types must be integers.
2681 ``fptrunc (CST to TYPE)``
2682 Truncate a floating point constant to another floating point type.
2683 The size of CST must be larger than the size of TYPE. Both types
2684 must be floating point.
2685 ``fpext (CST to TYPE)``
2686 Floating point extend a constant to another type. The size of CST
2687 must be smaller or equal to the size of TYPE. Both types must be
2689 ``fptoui (CST to TYPE)``
2690 Convert a floating point constant to the corresponding unsigned
2691 integer constant. TYPE must be a scalar or vector integer type. CST
2692 must be of scalar or vector floating point type. Both CST and TYPE
2693 must be scalars, or vectors of the same number of elements. If the
2694 value won't fit in the integer type, the results are undefined.
2695 ``fptosi (CST to TYPE)``
2696 Convert a floating point constant to the corresponding signed
2697 integer constant. TYPE must be a scalar or vector integer type. CST
2698 must be of scalar or vector floating point type. Both CST and TYPE
2699 must be scalars, or vectors of the same number of elements. If the
2700 value won't fit in the integer type, the results are undefined.
2701 ``uitofp (CST to TYPE)``
2702 Convert an unsigned integer constant to the corresponding floating
2703 point constant. TYPE must be a scalar or vector floating point type.
2704 CST must be of scalar or vector integer type. Both CST and TYPE must
2705 be scalars, or vectors of the same number of elements. If the value
2706 won't fit in the floating point type, the results are undefined.
2707 ``sitofp (CST to TYPE)``
2708 Convert a signed integer constant to the corresponding floating
2709 point constant. TYPE must be a scalar or vector floating point type.
2710 CST must be of scalar or vector integer type. Both CST and TYPE must
2711 be scalars, or vectors of the same number of elements. If the value
2712 won't fit in the floating point type, the results are undefined.
2713 ``ptrtoint (CST to TYPE)``
2714 Convert a pointer typed constant to the corresponding integer
2715 constant. ``TYPE`` must be an integer type. ``CST`` must be of
2716 pointer type. The ``CST`` value is zero extended, truncated, or
2717 unchanged to make it fit in ``TYPE``.
2718 ``inttoptr (CST to TYPE)``
2719 Convert an integer constant to a pointer constant. TYPE must be a
2720 pointer type. CST must be of integer type. The CST value is zero
2721 extended, truncated, or unchanged to make it fit in a pointer size.
2722 This one is *really* dangerous!
2723 ``bitcast (CST to TYPE)``
2724 Convert a constant, CST, to another TYPE. The constraints of the
2725 operands are the same as those for the :ref:`bitcast
2726 instruction <i_bitcast>`.
2727 ``addrspacecast (CST to TYPE)``
2728 Convert a constant pointer or constant vector of pointer, CST, to another
2729 TYPE in a different address space. The constraints of the operands are the
2730 same as those for the :ref:`addrspacecast instruction <i_addrspacecast>`.
2731 ``getelementptr (TY, CSTPTR, IDX0, IDX1, ...)``, ``getelementptr inbounds (TY, CSTPTR, IDX0, IDX1, ...)``
2732 Perform the :ref:`getelementptr operation <i_getelementptr>` on
2733 constants. As with the :ref:`getelementptr <i_getelementptr>`
2734 instruction, the index list may have zero or more indexes, which are
2735 required to make sense for the type of "pointer to TY".
2736 ``select (COND, VAL1, VAL2)``
2737 Perform the :ref:`select operation <i_select>` on constants.
2738 ``icmp COND (VAL1, VAL2)``
2739 Performs the :ref:`icmp operation <i_icmp>` on constants.
2740 ``fcmp COND (VAL1, VAL2)``
2741 Performs the :ref:`fcmp operation <i_fcmp>` on constants.
2742 ``extractelement (VAL, IDX)``
2743 Perform the :ref:`extractelement operation <i_extractelement>` on
2745 ``insertelement (VAL, ELT, IDX)``
2746 Perform the :ref:`insertelement operation <i_insertelement>` on
2748 ``shufflevector (VEC1, VEC2, IDXMASK)``
2749 Perform the :ref:`shufflevector operation <i_shufflevector>` on
2751 ``extractvalue (VAL, IDX0, IDX1, ...)``
2752 Perform the :ref:`extractvalue operation <i_extractvalue>` on
2753 constants. The index list is interpreted in a similar manner as
2754 indices in a ':ref:`getelementptr <i_getelementptr>`' operation. At
2755 least one index value must be specified.
2756 ``insertvalue (VAL, ELT, IDX0, IDX1, ...)``
2757 Perform the :ref:`insertvalue operation <i_insertvalue>` on constants.
2758 The index list is interpreted in a similar manner as indices in a
2759 ':ref:`getelementptr <i_getelementptr>`' operation. At least one index
2760 value must be specified.
2761 ``OPCODE (LHS, RHS)``
2762 Perform the specified operation of the LHS and RHS constants. OPCODE
2763 may be any of the :ref:`binary <binaryops>` or :ref:`bitwise
2764 binary <bitwiseops>` operations. The constraints on operands are
2765 the same as those for the corresponding instruction (e.g. no bitwise
2766 operations on floating point values are allowed).
2773 Inline Assembler Expressions
2774 ----------------------------
2776 LLVM supports inline assembler expressions (as opposed to :ref:`Module-Level
2777 Inline Assembly <moduleasm>`) through the use of a special value. This
2778 value represents the inline assembler as a string (containing the
2779 instructions to emit), a list of operand constraints (stored as a
2780 string), a flag that indicates whether or not the inline asm expression
2781 has side effects, and a flag indicating whether the function containing
2782 the asm needs to align its stack conservatively. An example inline
2783 assembler expression is:
2785 .. code-block:: llvm
2787 i32 (i32) asm "bswap $0", "=r,r"
2789 Inline assembler expressions may **only** be used as the callee operand
2790 of a :ref:`call <i_call>` or an :ref:`invoke <i_invoke>` instruction.
2791 Thus, typically we have:
2793 .. code-block:: llvm
2795 %X = call i32 asm "bswap $0", "=r,r"(i32 %Y)
2797 Inline asms with side effects not visible in the constraint list must be
2798 marked as having side effects. This is done through the use of the
2799 '``sideeffect``' keyword, like so:
2801 .. code-block:: llvm
2803 call void asm sideeffect "eieio", ""()
2805 In some cases inline asms will contain code that will not work unless
2806 the stack is aligned in some way, such as calls or SSE instructions on
2807 x86, yet will not contain code that does that alignment within the asm.
2808 The compiler should make conservative assumptions about what the asm
2809 might contain and should generate its usual stack alignment code in the
2810 prologue if the '``alignstack``' keyword is present:
2812 .. code-block:: llvm
2814 call void asm alignstack "eieio", ""()
2816 Inline asms also support using non-standard assembly dialects. The
2817 assumed dialect is ATT. When the '``inteldialect``' keyword is present,
2818 the inline asm is using the Intel dialect. Currently, ATT and Intel are
2819 the only supported dialects. An example is:
2821 .. code-block:: llvm
2823 call void asm inteldialect "eieio", ""()
2825 If multiple keywords appear the '``sideeffect``' keyword must come
2826 first, the '``alignstack``' keyword second and the '``inteldialect``'
2832 The call instructions that wrap inline asm nodes may have a
2833 "``!srcloc``" MDNode attached to it that contains a list of constant
2834 integers. If present, the code generator will use the integer as the
2835 location cookie value when report errors through the ``LLVMContext``
2836 error reporting mechanisms. This allows a front-end to correlate backend
2837 errors that occur with inline asm back to the source code that produced
2840 .. code-block:: llvm
2842 call void asm sideeffect "something bad", ""(), !srcloc !42
2844 !42 = !{ i32 1234567 }
2846 It is up to the front-end to make sense of the magic numbers it places
2847 in the IR. If the MDNode contains multiple constants, the code generator
2848 will use the one that corresponds to the line of the asm that the error
2856 LLVM IR allows metadata to be attached to instructions in the program
2857 that can convey extra information about the code to the optimizers and
2858 code generator. One example application of metadata is source-level
2859 debug information. There are two metadata primitives: strings and nodes.
2861 Metadata does not have a type, and is not a value. If referenced from a
2862 ``call`` instruction, it uses the ``metadata`` type.
2864 All metadata are identified in syntax by a exclamation point ('``!``').
2866 .. _metadata-string:
2868 Metadata Nodes and Metadata Strings
2869 -----------------------------------
2871 A metadata string is a string surrounded by double quotes. It can
2872 contain any character by escaping non-printable characters with
2873 "``\xx``" where "``xx``" is the two digit hex code. For example:
2876 Metadata nodes are represented with notation similar to structure
2877 constants (a comma separated list of elements, surrounded by braces and
2878 preceded by an exclamation point). Metadata nodes can have any values as
2879 their operand. For example:
2881 .. code-block:: llvm
2883 !{ !"test\00", i32 10}
2885 Metadata nodes that aren't uniqued use the ``distinct`` keyword. For example:
2887 .. code-block:: llvm
2889 !0 = distinct !{!"test\00", i32 10}
2891 ``distinct`` nodes are useful when nodes shouldn't be merged based on their
2892 content. They can also occur when transformations cause uniquing collisions
2893 when metadata operands change.
2895 A :ref:`named metadata <namedmetadatastructure>` is a collection of
2896 metadata nodes, which can be looked up in the module symbol table. For
2899 .. code-block:: llvm
2903 Metadata can be used as function arguments. Here ``llvm.dbg.value``
2904 function is using two metadata arguments:
2906 .. code-block:: llvm
2908 call void @llvm.dbg.value(metadata !24, i64 0, metadata !25)
2910 Metadata can be attached with an instruction. Here metadata ``!21`` is
2911 attached to the ``add`` instruction using the ``!dbg`` identifier:
2913 .. code-block:: llvm
2915 %indvar.next = add i64 %indvar, 1, !dbg !21
2917 More information about specific metadata nodes recognized by the
2918 optimizers and code generator is found below.
2920 .. _specialized-metadata:
2922 Specialized Metadata Nodes
2923 ^^^^^^^^^^^^^^^^^^^^^^^^^^
2925 Specialized metadata nodes are custom data structures in metadata (as opposed
2926 to generic tuples). Their fields are labelled, and can be specified in any
2929 These aren't inherently debug info centric, but currently all the specialized
2930 metadata nodes are related to debug info.
2937 ``DICompileUnit`` nodes represent a compile unit. The ``enums:``,
2938 ``retainedTypes:``, ``subprograms:``, ``globals:`` and ``imports:`` fields are
2939 tuples containing the debug info to be emitted along with the compile unit,
2940 regardless of code optimizations (some nodes are only emitted if there are
2941 references to them from instructions).
2943 .. code-block:: llvm
2945 !0 = !DICompileUnit(language: DW_LANG_C99, file: !1, producer: "clang",
2946 isOptimized: true, flags: "-O2", runtimeVersion: 2,
2947 splitDebugFilename: "abc.debug", emissionKind: 1,
2948 enums: !2, retainedTypes: !3, subprograms: !4,
2949 globals: !5, imports: !6)
2951 Compile unit descriptors provide the root scope for objects declared in a
2952 specific compilation unit. File descriptors are defined using this scope.
2953 These descriptors are collected by a named metadata ``!llvm.dbg.cu``. They
2954 keep track of subprograms, global variables, type information, and imported
2955 entities (declarations and namespaces).
2962 ``DIFile`` nodes represent files. The ``filename:`` can include slashes.
2964 .. code-block:: llvm
2966 !0 = !DIFile(filename: "path/to/file", directory: "/path/to/dir")
2968 Files are sometimes used in ``scope:`` fields, and are the only valid target
2969 for ``file:`` fields.
2976 ``DIBasicType`` nodes represent primitive types, such as ``int``, ``bool`` and
2977 ``float``. ``tag:`` defaults to ``DW_TAG_base_type``.
2979 .. code-block:: llvm
2981 !0 = !DIBasicType(name: "unsigned char", size: 8, align: 8,
2982 encoding: DW_ATE_unsigned_char)
2983 !1 = !DIBasicType(tag: DW_TAG_unspecified_type, name: "decltype(nullptr)")
2985 The ``encoding:`` describes the details of the type. Usually it's one of the
2988 .. code-block:: llvm
2994 DW_ATE_signed_char = 6
2996 DW_ATE_unsigned_char = 8
2998 .. _DISubroutineType:
3003 ``DISubroutineType`` nodes represent subroutine types. Their ``types:`` field
3004 refers to a tuple; the first operand is the return type, while the rest are the
3005 types of the formal arguments in order. If the first operand is ``null``, that
3006 represents a function with no return value (such as ``void foo() {}`` in C++).
3008 .. code-block:: llvm
3010 !0 = !BasicType(name: "int", size: 32, align: 32, DW_ATE_signed)
3011 !1 = !BasicType(name: "char", size: 8, align: 8, DW_ATE_signed_char)
3012 !2 = !DISubroutineType(types: !{null, !0, !1}) ; void (int, char)
3019 ``DIDerivedType`` nodes represent types derived from other types, such as
3022 .. code-block:: llvm
3024 !0 = !DIBasicType(name: "unsigned char", size: 8, align: 8,
3025 encoding: DW_ATE_unsigned_char)
3026 !1 = !DIDerivedType(tag: DW_TAG_pointer_type, baseType: !0, size: 32,
3029 The following ``tag:`` values are valid:
3031 .. code-block:: llvm
3033 DW_TAG_formal_parameter = 5
3035 DW_TAG_pointer_type = 15
3036 DW_TAG_reference_type = 16
3038 DW_TAG_ptr_to_member_type = 31
3039 DW_TAG_const_type = 38
3040 DW_TAG_volatile_type = 53
3041 DW_TAG_restrict_type = 55
3043 ``DW_TAG_member`` is used to define a member of a :ref:`composite type
3044 <DICompositeType>` or :ref:`subprogram <DISubprogram>`. The type of the member
3045 is the ``baseType:``. The ``offset:`` is the member's bit offset.
3046 ``DW_TAG_formal_parameter`` is used to define a member which is a formal
3047 argument of a subprogram.
3049 ``DW_TAG_typedef`` is used to provide a name for the ``baseType:``.
3051 ``DW_TAG_pointer_type``, ``DW_TAG_reference_type``, ``DW_TAG_const_type``,
3052 ``DW_TAG_volatile_type`` and ``DW_TAG_restrict_type`` are used to qualify the
3055 Note that the ``void *`` type is expressed as a type derived from NULL.
3057 .. _DICompositeType:
3062 ``DICompositeType`` nodes represent types composed of other types, like
3063 structures and unions. ``elements:`` points to a tuple of the composed types.
3065 If the source language supports ODR, the ``identifier:`` field gives the unique
3066 identifier used for type merging between modules. When specified, other types
3067 can refer to composite types indirectly via a :ref:`metadata string
3068 <metadata-string>` that matches their identifier.
3070 .. code-block:: llvm
3072 !0 = !DIEnumerator(name: "SixKind", value: 7)
3073 !1 = !DIEnumerator(name: "SevenKind", value: 7)
3074 !2 = !DIEnumerator(name: "NegEightKind", value: -8)
3075 !3 = !DICompositeType(tag: DW_TAG_enumeration_type, name: "Enum", file: !12,
3076 line: 2, size: 32, align: 32, identifier: "_M4Enum",
3077 elements: !{!0, !1, !2})
3079 The following ``tag:`` values are valid:
3081 .. code-block:: llvm
3083 DW_TAG_array_type = 1
3084 DW_TAG_class_type = 2
3085 DW_TAG_enumeration_type = 4
3086 DW_TAG_structure_type = 19
3087 DW_TAG_union_type = 23
3088 DW_TAG_subroutine_type = 21
3089 DW_TAG_inheritance = 28
3092 For ``DW_TAG_array_type``, the ``elements:`` should be :ref:`subrange
3093 descriptors <DISubrange>`, each representing the range of subscripts at that
3094 level of indexing. The ``DIFlagVector`` flag to ``flags:`` indicates that an
3095 array type is a native packed vector.
3097 For ``DW_TAG_enumeration_type``, the ``elements:`` should be :ref:`enumerator
3098 descriptors <DIEnumerator>`, each representing the definition of an enumeration
3099 value for the set. All enumeration type descriptors are collected in the
3100 ``enums:`` field of the :ref:`compile unit <DICompileUnit>`.
3102 For ``DW_TAG_structure_type``, ``DW_TAG_class_type``, and
3103 ``DW_TAG_union_type``, the ``elements:`` should be :ref:`derived types
3104 <DIDerivedType>` with ``tag: DW_TAG_member`` or ``tag: DW_TAG_inheritance``.
3111 ``DISubrange`` nodes are the elements for ``DW_TAG_array_type`` variants of
3112 :ref:`DICompositeType`. ``count: -1`` indicates an empty array.
3114 .. code-block:: llvm
3116 !0 = !DISubrange(count: 5, lowerBound: 0) ; array counting from 0
3117 !1 = !DISubrange(count: 5, lowerBound: 1) ; array counting from 1
3118 !2 = !DISubrange(count: -1) ; empty array.
3125 ``DIEnumerator`` nodes are the elements for ``DW_TAG_enumeration_type``
3126 variants of :ref:`DICompositeType`.
3128 .. code-block:: llvm
3130 !0 = !DIEnumerator(name: "SixKind", value: 7)
3131 !1 = !DIEnumerator(name: "SevenKind", value: 7)
3132 !2 = !DIEnumerator(name: "NegEightKind", value: -8)
3134 DITemplateTypeParameter
3135 """""""""""""""""""""""
3137 ``DITemplateTypeParameter`` nodes represent type parameters to generic source
3138 language constructs. They are used (optionally) in :ref:`DICompositeType` and
3139 :ref:`DISubprogram` ``templateParams:`` fields.
3141 .. code-block:: llvm
3143 !0 = !DITemplateTypeParameter(name: "Ty", type: !1)
3145 DITemplateValueParameter
3146 """"""""""""""""""""""""
3148 ``DITemplateValueParameter`` nodes represent value parameters to generic source
3149 language constructs. ``tag:`` defaults to ``DW_TAG_template_value_parameter``,
3150 but if specified can also be set to ``DW_TAG_GNU_template_template_param`` or
3151 ``DW_TAG_GNU_template_param_pack``. They are used (optionally) in
3152 :ref:`DICompositeType` and :ref:`DISubprogram` ``templateParams:`` fields.
3154 .. code-block:: llvm
3156 !0 = !DITemplateValueParameter(name: "Ty", type: !1, value: i32 7)
3161 ``DINamespace`` nodes represent namespaces in the source language.
3163 .. code-block:: llvm
3165 !0 = !DINamespace(name: "myawesomeproject", scope: !1, file: !2, line: 7)
3170 ``DIGlobalVariable`` nodes represent global variables in the source language.
3172 .. code-block:: llvm
3174 !0 = !DIGlobalVariable(name: "foo", linkageName: "foo", scope: !1,
3175 file: !2, line: 7, type: !3, isLocal: true,
3176 isDefinition: false, variable: i32* @foo,
3179 All global variables should be referenced by the `globals:` field of a
3180 :ref:`compile unit <DICompileUnit>`.
3187 ``DISubprogram`` nodes represent functions from the source language. The
3188 ``variables:`` field points at :ref:`variables <DILocalVariable>` that must be
3189 retained, even if their IR counterparts are optimized out of the IR. The
3190 ``type:`` field must point at an :ref:`DISubroutineType`.
3192 .. code-block:: llvm
3194 !0 = !DISubprogram(name: "foo", linkageName: "_Zfoov", scope: !1,
3195 file: !2, line: 7, type: !3, isLocal: true,
3196 isDefinition: false, scopeLine: 8, containingType: !4,
3197 virtuality: DW_VIRTUALITY_pure_virtual, virtualIndex: 10,
3198 flags: DIFlagPrototyped, isOptimized: true,
3199 function: void ()* @_Z3foov,
3200 templateParams: !5, declaration: !6, variables: !7)
3207 ``DILexicalBlock`` nodes describe nested blocks within a :ref:`subprogram
3208 <DISubprogram>`. The line number and column numbers are used to dinstinguish
3209 two lexical blocks at same depth. They are valid targets for ``scope:``
3212 .. code-block:: llvm
3214 !0 = distinct !DILexicalBlock(scope: !1, file: !2, line: 7, column: 35)
3216 Usually lexical blocks are ``distinct`` to prevent node merging based on
3219 .. _DILexicalBlockFile:
3224 ``DILexicalBlockFile`` nodes are used to discriminate between sections of a
3225 :ref:`lexical block <DILexicalBlock>`. The ``file:`` field can be changed to
3226 indicate textual inclusion, or the ``discriminator:`` field can be used to
3227 discriminate between control flow within a single block in the source language.
3229 .. code-block:: llvm
3231 !0 = !DILexicalBlock(scope: !3, file: !4, line: 7, column: 35)
3232 !1 = !DILexicalBlockFile(scope: !0, file: !4, discriminator: 0)
3233 !2 = !DILexicalBlockFile(scope: !0, file: !4, discriminator: 1)
3240 ``DILocation`` nodes represent source debug locations. The ``scope:`` field is
3241 mandatory, and points at an :ref:`DILexicalBlockFile`, an
3242 :ref:`DILexicalBlock`, or an :ref:`DISubprogram`.
3244 .. code-block:: llvm
3246 !0 = !DILocation(line: 2900, column: 42, scope: !1, inlinedAt: !2)
3248 .. _DILocalVariable:
3253 ``DILocalVariable`` nodes represent local variables in the source language.
3254 Instead of ``DW_TAG_variable``, they use LLVM-specific fake tags to
3255 discriminate between local variables (``DW_TAG_auto_variable``) and subprogram
3256 arguments (``DW_TAG_arg_variable``). In the latter case, the ``arg:`` field
3257 specifies the argument position, and this variable will be included in the
3258 ``variables:`` field of its :ref:`DISubprogram`.
3260 .. code-block:: llvm
3262 !0 = !DILocalVariable(tag: DW_TAG_arg_variable, name: "this", arg: 0,
3263 scope: !3, file: !2, line: 7, type: !3,
3264 flags: DIFlagArtificial)
3265 !1 = !DILocalVariable(tag: DW_TAG_arg_variable, name: "x", arg: 1,
3266 scope: !4, file: !2, line: 7, type: !3)
3267 !1 = !DILocalVariable(tag: DW_TAG_auto_variable, name: "y",
3268 scope: !5, file: !2, line: 7, type: !3)
3273 ``DIExpression`` nodes represent DWARF expression sequences. They are used in
3274 :ref:`debug intrinsics<dbg_intrinsics>` (such as ``llvm.dbg.declare``) to
3275 describe how the referenced LLVM variable relates to the source language
3278 The current supported vocabulary is limited:
3280 - ``DW_OP_deref`` dereferences the working expression.
3281 - ``DW_OP_plus, 93`` adds ``93`` to the working expression.
3282 - ``DW_OP_bit_piece, 16, 8`` specifies the offset and size (``16`` and ``8``
3283 here, respectively) of the variable piece from the working expression.
3285 .. code-block:: llvm
3287 !0 = !DIExpression(DW_OP_deref)
3288 !1 = !DIExpression(DW_OP_plus, 3)
3289 !2 = !DIExpression(DW_OP_bit_piece, 3, 7)
3290 !3 = !DIExpression(DW_OP_deref, DW_OP_plus, 3, DW_OP_bit_piece, 3, 7)
3295 ``DIObjCProperty`` nodes represent Objective-C property nodes.
3297 .. code-block:: llvm
3299 !3 = !DIObjCProperty(name: "foo", file: !1, line: 7, setter: "setFoo",
3300 getter: "getFoo", attributes: 7, type: !2)
3305 ``DIImportedEntity`` nodes represent entities (such as modules) imported into a
3308 .. code-block:: llvm
3310 !2 = !DIImportedEntity(tag: DW_TAG_imported_module, name: "foo", scope: !0,
3311 entity: !1, line: 7)
3316 In LLVM IR, memory does not have types, so LLVM's own type system is not
3317 suitable for doing TBAA. Instead, metadata is added to the IR to
3318 describe a type system of a higher level language. This can be used to
3319 implement typical C/C++ TBAA, but it can also be used to implement
3320 custom alias analysis behavior for other languages.
3322 The current metadata format is very simple. TBAA metadata nodes have up
3323 to three fields, e.g.:
3325 .. code-block:: llvm
3327 !0 = !{ !"an example type tree" }
3328 !1 = !{ !"int", !0 }
3329 !2 = !{ !"float", !0 }
3330 !3 = !{ !"const float", !2, i64 1 }
3332 The first field is an identity field. It can be any value, usually a
3333 metadata string, which uniquely identifies the type. The most important
3334 name in the tree is the name of the root node. Two trees with different
3335 root node names are entirely disjoint, even if they have leaves with
3338 The second field identifies the type's parent node in the tree, or is
3339 null or omitted for a root node. A type is considered to alias all of
3340 its descendants and all of its ancestors in the tree. Also, a type is
3341 considered to alias all types in other trees, so that bitcode produced
3342 from multiple front-ends is handled conservatively.
3344 If the third field is present, it's an integer which if equal to 1
3345 indicates that the type is "constant" (meaning
3346 ``pointsToConstantMemory`` should return true; see `other useful
3347 AliasAnalysis methods <AliasAnalysis.html#OtherItfs>`_).
3349 '``tbaa.struct``' Metadata
3350 ^^^^^^^^^^^^^^^^^^^^^^^^^^
3352 The :ref:`llvm.memcpy <int_memcpy>` is often used to implement
3353 aggregate assignment operations in C and similar languages, however it
3354 is defined to copy a contiguous region of memory, which is more than
3355 strictly necessary for aggregate types which contain holes due to
3356 padding. Also, it doesn't contain any TBAA information about the fields
3359 ``!tbaa.struct`` metadata can describe which memory subregions in a
3360 memcpy are padding and what the TBAA tags of the struct are.
3362 The current metadata format is very simple. ``!tbaa.struct`` metadata
3363 nodes are a list of operands which are in conceptual groups of three.
3364 For each group of three, the first operand gives the byte offset of a
3365 field in bytes, the second gives its size in bytes, and the third gives
3368 .. code-block:: llvm
3370 !4 = !{ i64 0, i64 4, !1, i64 8, i64 4, !2 }
3372 This describes a struct with two fields. The first is at offset 0 bytes
3373 with size 4 bytes, and has tbaa tag !1. The second is at offset 8 bytes
3374 and has size 4 bytes and has tbaa tag !2.
3376 Note that the fields need not be contiguous. In this example, there is a
3377 4 byte gap between the two fields. This gap represents padding which
3378 does not carry useful data and need not be preserved.
3380 '``noalias``' and '``alias.scope``' Metadata
3381 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
3383 ``noalias`` and ``alias.scope`` metadata provide the ability to specify generic
3384 noalias memory-access sets. This means that some collection of memory access
3385 instructions (loads, stores, memory-accessing calls, etc.) that carry
3386 ``noalias`` metadata can specifically be specified not to alias with some other
3387 collection of memory access instructions that carry ``alias.scope`` metadata.
3388 Each type of metadata specifies a list of scopes where each scope has an id and
3389 a domain. When evaluating an aliasing query, if for some domain, the set
3390 of scopes with that domain in one instruction's ``alias.scope`` list is a
3391 subset of (or equal to) the set of scopes for that domain in another
3392 instruction's ``noalias`` list, then the two memory accesses are assumed not to
3395 The metadata identifying each domain is itself a list containing one or two
3396 entries. The first entry is the name of the domain. Note that if the name is a
3397 string then it can be combined accross functions and translation units. A
3398 self-reference can be used to create globally unique domain names. A
3399 descriptive string may optionally be provided as a second list entry.
3401 The metadata identifying each scope is also itself a list containing two or
3402 three entries. The first entry is the name of the scope. Note that if the name
3403 is a string then it can be combined accross functions and translation units. A
3404 self-reference can be used to create globally unique scope names. A metadata
3405 reference to the scope's domain is the second entry. A descriptive string may
3406 optionally be provided as a third list entry.
3410 .. code-block:: llvm
3412 ; Two scope domains:
3416 ; Some scopes in these domains:
3422 !5 = !{!4} ; A list containing only scope !4
3426 ; These two instructions don't alias:
3427 %0 = load float, float* %c, align 4, !alias.scope !5
3428 store float %0, float* %arrayidx.i, align 4, !noalias !5
3430 ; These two instructions also don't alias (for domain !1, the set of scopes
3431 ; in the !alias.scope equals that in the !noalias list):
3432 %2 = load float, float* %c, align 4, !alias.scope !5
3433 store float %2, float* %arrayidx.i2, align 4, !noalias !6
3435 ; These two instructions may alias (for domain !0, the set of scopes in
3436 ; the !noalias list is not a superset of, or equal to, the scopes in the
3437 ; !alias.scope list):
3438 %2 = load float, float* %c, align 4, !alias.scope !6
3439 store float %0, float* %arrayidx.i, align 4, !noalias !7
3441 '``fpmath``' Metadata
3442 ^^^^^^^^^^^^^^^^^^^^^
3444 ``fpmath`` metadata may be attached to any instruction of floating point
3445 type. It can be used to express the maximum acceptable error in the
3446 result of that instruction, in ULPs, thus potentially allowing the
3447 compiler to use a more efficient but less accurate method of computing
3448 it. ULP is defined as follows:
3450 If ``x`` is a real number that lies between two finite consecutive
3451 floating-point numbers ``a`` and ``b``, without being equal to one
3452 of them, then ``ulp(x) = |b - a|``, otherwise ``ulp(x)`` is the
3453 distance between the two non-equal finite floating-point numbers
3454 nearest ``x``. Moreover, ``ulp(NaN)`` is ``NaN``.
3456 The metadata node shall consist of a single positive floating point
3457 number representing the maximum relative error, for example:
3459 .. code-block:: llvm
3461 !0 = !{ float 2.5 } ; maximum acceptable inaccuracy is 2.5 ULPs
3465 '``range``' Metadata
3466 ^^^^^^^^^^^^^^^^^^^^
3468 ``range`` metadata may be attached only to ``load``, ``call`` and ``invoke`` of
3469 integer types. It expresses the possible ranges the loaded value or the value
3470 returned by the called function at this call site is in. The ranges are
3471 represented with a flattened list of integers. The loaded value or the value
3472 returned is known to be in the union of the ranges defined by each consecutive
3473 pair. Each pair has the following properties:
3475 - The type must match the type loaded by the instruction.
3476 - The pair ``a,b`` represents the range ``[a,b)``.
3477 - Both ``a`` and ``b`` are constants.
3478 - The range is allowed to wrap.
3479 - The range should not represent the full or empty set. That is,
3482 In addition, the pairs must be in signed order of the lower bound and
3483 they must be non-contiguous.
3487 .. code-block:: llvm
3489 %a = load i8, i8* %x, align 1, !range !0 ; Can only be 0 or 1
3490 %b = load i8, i8* %y, align 1, !range !1 ; Can only be 255 (-1), 0 or 1
3491 %c = call i8 @foo(), !range !2 ; Can only be 0, 1, 3, 4 or 5
3492 %d = invoke i8 @bar() to label %cont
3493 unwind label %lpad, !range !3 ; Can only be -2, -1, 3, 4 or 5
3495 !0 = !{ i8 0, i8 2 }
3496 !1 = !{ i8 255, i8 2 }
3497 !2 = !{ i8 0, i8 2, i8 3, i8 6 }
3498 !3 = !{ i8 -2, i8 0, i8 3, i8 6 }
3503 It is sometimes useful to attach information to loop constructs. Currently,
3504 loop metadata is implemented as metadata attached to the branch instruction
3505 in the loop latch block. This type of metadata refer to a metadata node that is
3506 guaranteed to be separate for each loop. The loop identifier metadata is
3507 specified with the name ``llvm.loop``.
3509 The loop identifier metadata is implemented using a metadata that refers to
3510 itself to avoid merging it with any other identifier metadata, e.g.,
3511 during module linkage or function inlining. That is, each loop should refer
3512 to their own identification metadata even if they reside in separate functions.
3513 The following example contains loop identifier metadata for two separate loop
3516 .. code-block:: llvm
3521 The loop identifier metadata can be used to specify additional
3522 per-loop metadata. Any operands after the first operand can be treated
3523 as user-defined metadata. For example the ``llvm.loop.unroll.count``
3524 suggests an unroll factor to the loop unroller:
3526 .. code-block:: llvm
3528 br i1 %exitcond, label %._crit_edge, label %.lr.ph, !llvm.loop !0
3531 !1 = !{!"llvm.loop.unroll.count", i32 4}
3533 '``llvm.loop.vectorize``' and '``llvm.loop.interleave``'
3534 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
3536 Metadata prefixed with ``llvm.loop.vectorize`` or ``llvm.loop.interleave`` are
3537 used to control per-loop vectorization and interleaving parameters such as
3538 vectorization width and interleave count. These metadata should be used in
3539 conjunction with ``llvm.loop`` loop identification metadata. The
3540 ``llvm.loop.vectorize`` and ``llvm.loop.interleave`` metadata are only
3541 optimization hints and the optimizer will only interleave and vectorize loops if
3542 it believes it is safe to do so. The ``llvm.mem.parallel_loop_access`` metadata
3543 which contains information about loop-carried memory dependencies can be helpful
3544 in determining the safety of these transformations.
3546 '``llvm.loop.interleave.count``' Metadata
3547 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
3549 This metadata suggests an interleave count to the loop interleaver.
3550 The first operand is the string ``llvm.loop.interleave.count`` and the
3551 second operand is an integer specifying the interleave count. For
3554 .. code-block:: llvm
3556 !0 = !{!"llvm.loop.interleave.count", i32 4}
3558 Note that setting ``llvm.loop.interleave.count`` to 1 disables interleaving
3559 multiple iterations of the loop. If ``llvm.loop.interleave.count`` is set to 0
3560 then the interleave count will be determined automatically.
3562 '``llvm.loop.vectorize.enable``' Metadata
3563 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
3565 This metadata selectively enables or disables vectorization for the loop. The
3566 first operand is the string ``llvm.loop.vectorize.enable`` and the second operand
3567 is a bit. If the bit operand value is 1 vectorization is enabled. A value of
3568 0 disables vectorization:
3570 .. code-block:: llvm
3572 !0 = !{!"llvm.loop.vectorize.enable", i1 0}
3573 !1 = !{!"llvm.loop.vectorize.enable", i1 1}
3575 '``llvm.loop.vectorize.width``' Metadata
3576 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
3578 This metadata sets the target width of the vectorizer. The first
3579 operand is the string ``llvm.loop.vectorize.width`` and the second
3580 operand is an integer specifying the width. For example:
3582 .. code-block:: llvm
3584 !0 = !{!"llvm.loop.vectorize.width", i32 4}
3586 Note that setting ``llvm.loop.vectorize.width`` to 1 disables
3587 vectorization of the loop. If ``llvm.loop.vectorize.width`` is set to
3588 0 or if the loop does not have this metadata the width will be
3589 determined automatically.
3591 '``llvm.loop.unroll``'
3592 ^^^^^^^^^^^^^^^^^^^^^^
3594 Metadata prefixed with ``llvm.loop.unroll`` are loop unrolling
3595 optimization hints such as the unroll factor. ``llvm.loop.unroll``
3596 metadata should be used in conjunction with ``llvm.loop`` loop
3597 identification metadata. The ``llvm.loop.unroll`` metadata are only
3598 optimization hints and the unrolling will only be performed if the
3599 optimizer believes it is safe to do so.
3601 '``llvm.loop.unroll.count``' Metadata
3602 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
3604 This metadata suggests an unroll factor to the loop unroller. The
3605 first operand is the string ``llvm.loop.unroll.count`` and the second
3606 operand is a positive integer specifying the unroll factor. For
3609 .. code-block:: llvm
3611 !0 = !{!"llvm.loop.unroll.count", i32 4}
3613 If the trip count of the loop is less than the unroll count the loop
3614 will be partially unrolled.
3616 '``llvm.loop.unroll.disable``' Metadata
3617 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
3619 This metadata either disables loop unrolling. The metadata has a single operand
3620 which is the string ``llvm.loop.unroll.disable``. For example:
3622 .. code-block:: llvm
3624 !0 = !{!"llvm.loop.unroll.disable"}
3626 '``llvm.loop.unroll.runtime.disable``' Metadata
3627 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
3629 This metadata either disables runtime loop unrolling. The metadata has a single
3630 operand which is the string ``llvm.loop.unroll.runtime.disable``. For example:
3632 .. code-block:: llvm
3634 !0 = !{!"llvm.loop.unroll.runtime.disable"}
3636 '``llvm.loop.unroll.full``' Metadata
3637 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
3639 This metadata either suggests that the loop should be unrolled fully. The
3640 metadata has a single operand which is the string ``llvm.loop.unroll.disable``.
3643 .. code-block:: llvm
3645 !0 = !{!"llvm.loop.unroll.full"}
3650 Metadata types used to annotate memory accesses with information helpful
3651 for optimizations are prefixed with ``llvm.mem``.
3653 '``llvm.mem.parallel_loop_access``' Metadata
3654 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
3656 The ``llvm.mem.parallel_loop_access`` metadata refers to a loop identifier,
3657 or metadata containing a list of loop identifiers for nested loops.
3658 The metadata is attached to memory accessing instructions and denotes that
3659 no loop carried memory dependence exist between it and other instructions denoted
3660 with the same loop identifier.
3662 Precisely, given two instructions ``m1`` and ``m2`` that both have the
3663 ``llvm.mem.parallel_loop_access`` metadata, with ``L1`` and ``L2`` being the
3664 set of loops associated with that metadata, respectively, then there is no loop
3665 carried dependence between ``m1`` and ``m2`` for loops in both ``L1`` and
3668 As a special case, if all memory accessing instructions in a loop have
3669 ``llvm.mem.parallel_loop_access`` metadata that refers to that loop, then the
3670 loop has no loop carried memory dependences and is considered to be a parallel
3673 Note that if not all memory access instructions have such metadata referring to
3674 the loop, then the loop is considered not being trivially parallel. Additional
3675 memory dependence analysis is required to make that determination. As a fail
3676 safe mechanism, this causes loops that were originally parallel to be considered
3677 sequential (if optimization passes that are unaware of the parallel semantics
3678 insert new memory instructions into the loop body).
3680 Example of a loop that is considered parallel due to its correct use of
3681 both ``llvm.loop`` and ``llvm.mem.parallel_loop_access``
3682 metadata types that refer to the same loop identifier metadata.
3684 .. code-block:: llvm
3688 %val0 = load i32, i32* %arrayidx, !llvm.mem.parallel_loop_access !0
3690 store i32 %val0, i32* %arrayidx1, !llvm.mem.parallel_loop_access !0
3692 br i1 %exitcond, label %for.end, label %for.body, !llvm.loop !0
3698 It is also possible to have nested parallel loops. In that case the
3699 memory accesses refer to a list of loop identifier metadata nodes instead of
3700 the loop identifier metadata node directly:
3702 .. code-block:: llvm
3706 %val1 = load i32, i32* %arrayidx3, !llvm.mem.parallel_loop_access !2
3708 br label %inner.for.body
3712 %val0 = load i32, i32* %arrayidx1, !llvm.mem.parallel_loop_access !0
3714 store i32 %val0, i32* %arrayidx2, !llvm.mem.parallel_loop_access !0
3716 br i1 %exitcond, label %inner.for.end, label %inner.for.body, !llvm.loop !1
3720 store i32 %val1, i32* %arrayidx4, !llvm.mem.parallel_loop_access !2
3722 br i1 %exitcond, label %outer.for.end, label %outer.for.body, !llvm.loop !2
3724 outer.for.end: ; preds = %for.body
3726 !0 = !{!1, !2} ; a list of loop identifiers
3727 !1 = !{!1} ; an identifier for the inner loop
3728 !2 = !{!2} ; an identifier for the outer loop
3733 The ``llvm.bitsets`` global metadata is used to implement
3734 :doc:`bitsets <BitSets>`.
3736 Module Flags Metadata
3737 =====================
3739 Information about the module as a whole is difficult to convey to LLVM's
3740 subsystems. The LLVM IR isn't sufficient to transmit this information.
3741 The ``llvm.module.flags`` named metadata exists in order to facilitate
3742 this. These flags are in the form of key / value pairs --- much like a
3743 dictionary --- making it easy for any subsystem who cares about a flag to
3746 The ``llvm.module.flags`` metadata contains a list of metadata triplets.
3747 Each triplet has the following form:
3749 - The first element is a *behavior* flag, which specifies the behavior
3750 when two (or more) modules are merged together, and it encounters two
3751 (or more) metadata with the same ID. The supported behaviors are
3753 - The second element is a metadata string that is a unique ID for the
3754 metadata. Each module may only have one flag entry for each unique ID (not
3755 including entries with the **Require** behavior).
3756 - The third element is the value of the flag.
3758 When two (or more) modules are merged together, the resulting
3759 ``llvm.module.flags`` metadata is the union of the modules' flags. That is, for
3760 each unique metadata ID string, there will be exactly one entry in the merged
3761 modules ``llvm.module.flags`` metadata table, and the value for that entry will
3762 be determined by the merge behavior flag, as described below. The only exception
3763 is that entries with the *Require* behavior are always preserved.
3765 The following behaviors are supported:
3776 Emits an error if two values disagree, otherwise the resulting value
3777 is that of the operands.
3781 Emits a warning if two values disagree. The result value will be the
3782 operand for the flag from the first module being linked.
3786 Adds a requirement that another module flag be present and have a
3787 specified value after linking is performed. The value must be a
3788 metadata pair, where the first element of the pair is the ID of the
3789 module flag to be restricted, and the second element of the pair is
3790 the value the module flag should be restricted to. This behavior can
3791 be used to restrict the allowable results (via triggering of an
3792 error) of linking IDs with the **Override** behavior.
3796 Uses the specified value, regardless of the behavior or value of the
3797 other module. If both modules specify **Override**, but the values
3798 differ, an error will be emitted.
3802 Appends the two values, which are required to be metadata nodes.
3806 Appends the two values, which are required to be metadata
3807 nodes. However, duplicate entries in the second list are dropped
3808 during the append operation.
3810 It is an error for a particular unique flag ID to have multiple behaviors,
3811 except in the case of **Require** (which adds restrictions on another metadata
3812 value) or **Override**.
3814 An example of module flags:
3816 .. code-block:: llvm
3818 !0 = !{ i32 1, !"foo", i32 1 }
3819 !1 = !{ i32 4, !"bar", i32 37 }
3820 !2 = !{ i32 2, !"qux", i32 42 }
3821 !3 = !{ i32 3, !"qux",
3826 !llvm.module.flags = !{ !0, !1, !2, !3 }
3828 - Metadata ``!0`` has the ID ``!"foo"`` and the value '1'. The behavior
3829 if two or more ``!"foo"`` flags are seen is to emit an error if their
3830 values are not equal.
3832 - Metadata ``!1`` has the ID ``!"bar"`` and the value '37'. The
3833 behavior if two or more ``!"bar"`` flags are seen is to use the value
3836 - Metadata ``!2`` has the ID ``!"qux"`` and the value '42'. The
3837 behavior if two or more ``!"qux"`` flags are seen is to emit a
3838 warning if their values are not equal.
3840 - Metadata ``!3`` has the ID ``!"qux"`` and the value:
3846 The behavior is to emit an error if the ``llvm.module.flags`` does not
3847 contain a flag with the ID ``!"foo"`` that has the value '1' after linking is
3850 Objective-C Garbage Collection Module Flags Metadata
3851 ----------------------------------------------------
3853 On the Mach-O platform, Objective-C stores metadata about garbage
3854 collection in a special section called "image info". The metadata
3855 consists of a version number and a bitmask specifying what types of
3856 garbage collection are supported (if any) by the file. If two or more
3857 modules are linked together their garbage collection metadata needs to
3858 be merged rather than appended together.
3860 The Objective-C garbage collection module flags metadata consists of the
3861 following key-value pairs:
3870 * - ``Objective-C Version``
3871 - **[Required]** --- The Objective-C ABI version. Valid values are 1 and 2.
3873 * - ``Objective-C Image Info Version``
3874 - **[Required]** --- The version of the image info section. Currently
3877 * - ``Objective-C Image Info Section``
3878 - **[Required]** --- The section to place the metadata. Valid values are
3879 ``"__OBJC, __image_info, regular"`` for Objective-C ABI version 1, and
3880 ``"__DATA,__objc_imageinfo, regular, no_dead_strip"`` for
3881 Objective-C ABI version 2.
3883 * - ``Objective-C Garbage Collection``
3884 - **[Required]** --- Specifies whether garbage collection is supported or
3885 not. Valid values are 0, for no garbage collection, and 2, for garbage
3886 collection supported.
3888 * - ``Objective-C GC Only``
3889 - **[Optional]** --- Specifies that only garbage collection is supported.
3890 If present, its value must be 6. This flag requires that the
3891 ``Objective-C Garbage Collection`` flag have the value 2.
3893 Some important flag interactions:
3895 - If a module with ``Objective-C Garbage Collection`` set to 0 is
3896 merged with a module with ``Objective-C Garbage Collection`` set to
3897 2, then the resulting module has the
3898 ``Objective-C Garbage Collection`` flag set to 0.
3899 - A module with ``Objective-C Garbage Collection`` set to 0 cannot be
3900 merged with a module with ``Objective-C GC Only`` set to 6.
3902 Automatic Linker Flags Module Flags Metadata
3903 --------------------------------------------
3905 Some targets support embedding flags to the linker inside individual object
3906 files. Typically this is used in conjunction with language extensions which
3907 allow source files to explicitly declare the libraries they depend on, and have
3908 these automatically be transmitted to the linker via object files.
3910 These flags are encoded in the IR using metadata in the module flags section,
3911 using the ``Linker Options`` key. The merge behavior for this flag is required
3912 to be ``AppendUnique``, and the value for the key is expected to be a metadata
3913 node which should be a list of other metadata nodes, each of which should be a
3914 list of metadata strings defining linker options.
3916 For example, the following metadata section specifies two separate sets of
3917 linker options, presumably to link against ``libz`` and the ``Cocoa``
3920 !0 = !{ i32 6, !"Linker Options",
3923 !{ !"-framework", !"Cocoa" } } }
3924 !llvm.module.flags = !{ !0 }
3926 The metadata encoding as lists of lists of options, as opposed to a collapsed
3927 list of options, is chosen so that the IR encoding can use multiple option
3928 strings to specify e.g., a single library, while still having that specifier be
3929 preserved as an atomic element that can be recognized by a target specific
3930 assembly writer or object file emitter.
3932 Each individual option is required to be either a valid option for the target's
3933 linker, or an option that is reserved by the target specific assembly writer or
3934 object file emitter. No other aspect of these options is defined by the IR.
3936 C type width Module Flags Metadata
3937 ----------------------------------
3939 The ARM backend emits a section into each generated object file describing the
3940 options that it was compiled with (in a compiler-independent way) to prevent
3941 linking incompatible objects, and to allow automatic library selection. Some
3942 of these options are not visible at the IR level, namely wchar_t width and enum
3945 To pass this information to the backend, these options are encoded in module
3946 flags metadata, using the following key-value pairs:
3956 - * 0 --- sizeof(wchar_t) == 4
3957 * 1 --- sizeof(wchar_t) == 2
3960 - * 0 --- Enums are at least as large as an ``int``.
3961 * 1 --- Enums are stored in the smallest integer type which can
3962 represent all of its values.
3964 For example, the following metadata section specifies that the module was
3965 compiled with a ``wchar_t`` width of 4 bytes, and the underlying type of an
3966 enum is the smallest type which can represent all of its values::
3968 !llvm.module.flags = !{!0, !1}
3969 !0 = !{i32 1, !"short_wchar", i32 1}
3970 !1 = !{i32 1, !"short_enum", i32 0}
3972 .. _intrinsicglobalvariables:
3974 Intrinsic Global Variables
3975 ==========================
3977 LLVM has a number of "magic" global variables that contain data that
3978 affect code generation or other IR semantics. These are documented here.
3979 All globals of this sort should have a section specified as
3980 "``llvm.metadata``". This section and all globals that start with
3981 "``llvm.``" are reserved for use by LLVM.
3985 The '``llvm.used``' Global Variable
3986 -----------------------------------
3988 The ``@llvm.used`` global is an array which has
3989 :ref:`appending linkage <linkage_appending>`. This array contains a list of
3990 pointers to named global variables, functions and aliases which may optionally
3991 have a pointer cast formed of bitcast or getelementptr. For example, a legal
3994 .. code-block:: llvm
3999 @llvm.used = appending global [2 x i8*] [
4001 i8* bitcast (i32* @Y to i8*)
4002 ], section "llvm.metadata"
4004 If a symbol appears in the ``@llvm.used`` list, then the compiler, assembler,
4005 and linker are required to treat the symbol as if there is a reference to the
4006 symbol that it cannot see (which is why they have to be named). For example, if
4007 a variable has internal linkage and no references other than that from the
4008 ``@llvm.used`` list, it cannot be deleted. This is commonly used to represent
4009 references from inline asms and other things the compiler cannot "see", and
4010 corresponds to "``attribute((used))``" in GNU C.
4012 On some targets, the code generator must emit a directive to the
4013 assembler or object file to prevent the assembler and linker from
4014 molesting the symbol.
4016 .. _gv_llvmcompilerused:
4018 The '``llvm.compiler.used``' Global Variable
4019 --------------------------------------------
4021 The ``@llvm.compiler.used`` directive is the same as the ``@llvm.used``
4022 directive, except that it only prevents the compiler from touching the
4023 symbol. On targets that support it, this allows an intelligent linker to
4024 optimize references to the symbol without being impeded as it would be
4027 This is a rare construct that should only be used in rare circumstances,
4028 and should not be exposed to source languages.
4030 .. _gv_llvmglobalctors:
4032 The '``llvm.global_ctors``' Global Variable
4033 -------------------------------------------
4035 .. code-block:: llvm
4037 %0 = type { i32, void ()*, i8* }
4038 @llvm.global_ctors = appending global [1 x %0] [%0 { i32 65535, void ()* @ctor, i8* @data }]
4040 The ``@llvm.global_ctors`` array contains a list of constructor
4041 functions, priorities, and an optional associated global or function.
4042 The functions referenced by this array will be called in ascending order
4043 of priority (i.e. lowest first) when the module is loaded. The order of
4044 functions with the same priority is not defined.
4046 If the third field is present, non-null, and points to a global variable
4047 or function, the initializer function will only run if the associated
4048 data from the current module is not discarded.
4050 .. _llvmglobaldtors:
4052 The '``llvm.global_dtors``' Global Variable
4053 -------------------------------------------
4055 .. code-block:: llvm
4057 %0 = type { i32, void ()*, i8* }
4058 @llvm.global_dtors = appending global [1 x %0] [%0 { i32 65535, void ()* @dtor, i8* @data }]
4060 The ``@llvm.global_dtors`` array contains a list of destructor
4061 functions, priorities, and an optional associated global or function.
4062 The functions referenced by this array will be called in descending
4063 order of priority (i.e. highest first) when the module is unloaded. The
4064 order of functions with the same priority is not defined.
4066 If the third field is present, non-null, and points to a global variable
4067 or function, the destructor function will only run if the associated
4068 data from the current module is not discarded.
4070 Instruction Reference
4071 =====================
4073 The LLVM instruction set consists of several different classifications
4074 of instructions: :ref:`terminator instructions <terminators>`, :ref:`binary
4075 instructions <binaryops>`, :ref:`bitwise binary
4076 instructions <bitwiseops>`, :ref:`memory instructions <memoryops>`, and
4077 :ref:`other instructions <otherops>`.
4081 Terminator Instructions
4082 -----------------------
4084 As mentioned :ref:`previously <functionstructure>`, every basic block in a
4085 program ends with a "Terminator" instruction, which indicates which
4086 block should be executed after the current block is finished. These
4087 terminator instructions typically yield a '``void``' value: they produce
4088 control flow, not values (the one exception being the
4089 ':ref:`invoke <i_invoke>`' instruction).
4091 The terminator instructions are: ':ref:`ret <i_ret>`',
4092 ':ref:`br <i_br>`', ':ref:`switch <i_switch>`',
4093 ':ref:`indirectbr <i_indirectbr>`', ':ref:`invoke <i_invoke>`',
4094 ':ref:`resume <i_resume>`', and ':ref:`unreachable <i_unreachable>`'.
4098 '``ret``' Instruction
4099 ^^^^^^^^^^^^^^^^^^^^^
4106 ret <type> <value> ; Return a value from a non-void function
4107 ret void ; Return from void function
4112 The '``ret``' instruction is used to return control flow (and optionally
4113 a value) from a function back to the caller.
4115 There are two forms of the '``ret``' instruction: one that returns a
4116 value and then causes control flow, and one that just causes control
4122 The '``ret``' instruction optionally accepts a single argument, the
4123 return value. The type of the return value must be a ':ref:`first
4124 class <t_firstclass>`' type.
4126 A function is not :ref:`well formed <wellformed>` if it it has a non-void
4127 return type and contains a '``ret``' instruction with no return value or
4128 a return value with a type that does not match its type, or if it has a
4129 void return type and contains a '``ret``' instruction with a return
4135 When the '``ret``' instruction is executed, control flow returns back to
4136 the calling function's context. If the caller is a
4137 ":ref:`call <i_call>`" instruction, execution continues at the
4138 instruction after the call. If the caller was an
4139 ":ref:`invoke <i_invoke>`" instruction, execution continues at the
4140 beginning of the "normal" destination block. If the instruction returns
4141 a value, that value shall set the call or invoke instruction's return
4147 .. code-block:: llvm
4149 ret i32 5 ; Return an integer value of 5
4150 ret void ; Return from a void function
4151 ret { i32, i8 } { i32 4, i8 2 } ; Return a struct of values 4 and 2
4155 '``br``' Instruction
4156 ^^^^^^^^^^^^^^^^^^^^
4163 br i1 <cond>, label <iftrue>, label <iffalse>
4164 br label <dest> ; Unconditional branch
4169 The '``br``' instruction is used to cause control flow to transfer to a
4170 different basic block in the current function. There are two forms of
4171 this instruction, corresponding to a conditional branch and an
4172 unconditional branch.
4177 The conditional branch form of the '``br``' instruction takes a single
4178 '``i1``' value and two '``label``' values. The unconditional form of the
4179 '``br``' instruction takes a single '``label``' value as a target.
4184 Upon execution of a conditional '``br``' instruction, the '``i1``'
4185 argument is evaluated. If the value is ``true``, control flows to the
4186 '``iftrue``' ``label`` argument. If "cond" is ``false``, control flows
4187 to the '``iffalse``' ``label`` argument.
4192 .. code-block:: llvm
4195 %cond = icmp eq i32 %a, %b
4196 br i1 %cond, label %IfEqual, label %IfUnequal
4204 '``switch``' Instruction
4205 ^^^^^^^^^^^^^^^^^^^^^^^^
4212 switch <intty> <value>, label <defaultdest> [ <intty> <val>, label <dest> ... ]
4217 The '``switch``' instruction is used to transfer control flow to one of
4218 several different places. It is a generalization of the '``br``'
4219 instruction, allowing a branch to occur to one of many possible
4225 The '``switch``' instruction uses three parameters: an integer
4226 comparison value '``value``', a default '``label``' destination, and an
4227 array of pairs of comparison value constants and '``label``'s. The table
4228 is not allowed to contain duplicate constant entries.
4233 The ``switch`` instruction specifies a table of values and destinations.
4234 When the '``switch``' instruction is executed, this table is searched
4235 for the given value. If the value is found, control flow is transferred
4236 to the corresponding destination; otherwise, control flow is transferred
4237 to the default destination.
4242 Depending on properties of the target machine and the particular
4243 ``switch`` instruction, this instruction may be code generated in
4244 different ways. For example, it could be generated as a series of
4245 chained conditional branches or with a lookup table.
4250 .. code-block:: llvm
4252 ; Emulate a conditional br instruction
4253 %Val = zext i1 %value to i32
4254 switch i32 %Val, label %truedest [ i32 0, label %falsedest ]
4256 ; Emulate an unconditional br instruction
4257 switch i32 0, label %dest [ ]
4259 ; Implement a jump table:
4260 switch i32 %val, label %otherwise [ i32 0, label %onzero
4262 i32 2, label %ontwo ]
4266 '``indirectbr``' Instruction
4267 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4274 indirectbr <somety>* <address>, [ label <dest1>, label <dest2>, ... ]
4279 The '``indirectbr``' instruction implements an indirect branch to a
4280 label within the current function, whose address is specified by
4281 "``address``". Address must be derived from a
4282 :ref:`blockaddress <blockaddress>` constant.
4287 The '``address``' argument is the address of the label to jump to. The
4288 rest of the arguments indicate the full set of possible destinations
4289 that the address may point to. Blocks are allowed to occur multiple
4290 times in the destination list, though this isn't particularly useful.
4292 This destination list is required so that dataflow analysis has an
4293 accurate understanding of the CFG.
4298 Control transfers to the block specified in the address argument. All
4299 possible destination blocks must be listed in the label list, otherwise
4300 this instruction has undefined behavior. This implies that jumps to
4301 labels defined in other functions have undefined behavior as well.
4306 This is typically implemented with a jump through a register.
4311 .. code-block:: llvm
4313 indirectbr i8* %Addr, [ label %bb1, label %bb2, label %bb3 ]
4317 '``invoke``' Instruction
4318 ^^^^^^^^^^^^^^^^^^^^^^^^
4325 <result> = invoke [cconv] [ret attrs] <ptr to function ty> <function ptr val>(<function args>) [fn attrs]
4326 to label <normal label> unwind label <exception label>
4331 The '``invoke``' instruction causes control to transfer to a specified
4332 function, with the possibility of control flow transfer to either the
4333 '``normal``' label or the '``exception``' label. If the callee function
4334 returns with the "``ret``" instruction, control flow will return to the
4335 "normal" label. If the callee (or any indirect callees) returns via the
4336 ":ref:`resume <i_resume>`" instruction or other exception handling
4337 mechanism, control is interrupted and continued at the dynamically
4338 nearest "exception" label.
4340 The '``exception``' label is a `landing
4341 pad <ExceptionHandling.html#overview>`_ for the exception. As such,
4342 '``exception``' label is required to have the
4343 ":ref:`landingpad <i_landingpad>`" instruction, which contains the
4344 information about the behavior of the program after unwinding happens,
4345 as its first non-PHI instruction. The restrictions on the
4346 "``landingpad``" instruction's tightly couples it to the "``invoke``"
4347 instruction, so that the important information contained within the
4348 "``landingpad``" instruction can't be lost through normal code motion.
4353 This instruction requires several arguments:
4355 #. The optional "cconv" marker indicates which :ref:`calling
4356 convention <callingconv>` the call should use. If none is
4357 specified, the call defaults to using C calling conventions.
4358 #. The optional :ref:`Parameter Attributes <paramattrs>` list for return
4359 values. Only '``zeroext``', '``signext``', and '``inreg``' attributes
4361 #. '``ptr to function ty``': shall be the signature of the pointer to
4362 function value being invoked. In most cases, this is a direct
4363 function invocation, but indirect ``invoke``'s are just as possible,
4364 branching off an arbitrary pointer to function value.
4365 #. '``function ptr val``': An LLVM value containing a pointer to a
4366 function to be invoked.
4367 #. '``function args``': argument list whose types match the function
4368 signature argument types and parameter attributes. All arguments must
4369 be of :ref:`first class <t_firstclass>` type. If the function signature
4370 indicates the function accepts a variable number of arguments, the
4371 extra arguments can be specified.
4372 #. '``normal label``': the label reached when the called function
4373 executes a '``ret``' instruction.
4374 #. '``exception label``': the label reached when a callee returns via
4375 the :ref:`resume <i_resume>` instruction or other exception handling
4377 #. The optional :ref:`function attributes <fnattrs>` list. Only
4378 '``noreturn``', '``nounwind``', '``readonly``' and '``readnone``'
4379 attributes are valid here.
4384 This instruction is designed to operate as a standard '``call``'
4385 instruction in most regards. The primary difference is that it
4386 establishes an association with a label, which is used by the runtime
4387 library to unwind the stack.
4389 This instruction is used in languages with destructors to ensure that
4390 proper cleanup is performed in the case of either a ``longjmp`` or a
4391 thrown exception. Additionally, this is important for implementation of
4392 '``catch``' clauses in high-level languages that support them.
4394 For the purposes of the SSA form, the definition of the value returned
4395 by the '``invoke``' instruction is deemed to occur on the edge from the
4396 current block to the "normal" label. If the callee unwinds then no
4397 return value is available.
4402 .. code-block:: llvm
4404 %retval = invoke i32 @Test(i32 15) to label %Continue
4405 unwind label %TestCleanup ; i32:retval set
4406 %retval = invoke coldcc i32 %Testfnptr(i32 15) to label %Continue
4407 unwind label %TestCleanup ; i32:retval set
4411 '``resume``' Instruction
4412 ^^^^^^^^^^^^^^^^^^^^^^^^
4419 resume <type> <value>
4424 The '``resume``' instruction is a terminator instruction that has no
4430 The '``resume``' instruction requires one argument, which must have the
4431 same type as the result of any '``landingpad``' instruction in the same
4437 The '``resume``' instruction resumes propagation of an existing
4438 (in-flight) exception whose unwinding was interrupted with a
4439 :ref:`landingpad <i_landingpad>` instruction.
4444 .. code-block:: llvm
4446 resume { i8*, i32 } %exn
4450 '``unreachable``' Instruction
4451 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4463 The '``unreachable``' instruction has no defined semantics. This
4464 instruction is used to inform the optimizer that a particular portion of
4465 the code is not reachable. This can be used to indicate that the code
4466 after a no-return function cannot be reached, and other facts.
4471 The '``unreachable``' instruction has no defined semantics.
4478 Binary operators are used to do most of the computation in a program.
4479 They require two operands of the same type, execute an operation on
4480 them, and produce a single value. The operands might represent multiple
4481 data, as is the case with the :ref:`vector <t_vector>` data type. The
4482 result value has the same type as its operands.
4484 There are several different binary operators:
4488 '``add``' Instruction
4489 ^^^^^^^^^^^^^^^^^^^^^
4496 <result> = add <ty> <op1>, <op2> ; yields ty:result
4497 <result> = add nuw <ty> <op1>, <op2> ; yields ty:result
4498 <result> = add nsw <ty> <op1>, <op2> ; yields ty:result
4499 <result> = add nuw nsw <ty> <op1>, <op2> ; yields ty:result
4504 The '``add``' instruction returns the sum of its two operands.
4509 The two arguments to the '``add``' instruction must be
4510 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
4511 arguments must have identical types.
4516 The value produced is the integer sum of the two operands.
4518 If the sum has unsigned overflow, the result returned is the
4519 mathematical result modulo 2\ :sup:`n`\ , where n is the bit width of
4522 Because LLVM integers use a two's complement representation, this
4523 instruction is appropriate for both signed and unsigned integers.
4525 ``nuw`` and ``nsw`` stand for "No Unsigned Wrap" and "No Signed Wrap",
4526 respectively. If the ``nuw`` and/or ``nsw`` keywords are present, the
4527 result value of the ``add`` is a :ref:`poison value <poisonvalues>` if
4528 unsigned and/or signed overflow, respectively, occurs.
4533 .. code-block:: llvm
4535 <result> = add i32 4, %var ; yields i32:result = 4 + %var
4539 '``fadd``' Instruction
4540 ^^^^^^^^^^^^^^^^^^^^^^
4547 <result> = fadd [fast-math flags]* <ty> <op1>, <op2> ; yields ty:result
4552 The '``fadd``' instruction returns the sum of its two operands.
4557 The two arguments to the '``fadd``' instruction must be :ref:`floating
4558 point <t_floating>` or :ref:`vector <t_vector>` of floating point values.
4559 Both arguments must have identical types.
4564 The value produced is the floating point sum of the two operands. This
4565 instruction can also take any number of :ref:`fast-math flags <fastmath>`,
4566 which are optimization hints to enable otherwise unsafe floating point
4572 .. code-block:: llvm
4574 <result> = fadd float 4.0, %var ; yields float:result = 4.0 + %var
4576 '``sub``' Instruction
4577 ^^^^^^^^^^^^^^^^^^^^^
4584 <result> = sub <ty> <op1>, <op2> ; yields ty:result
4585 <result> = sub nuw <ty> <op1>, <op2> ; yields ty:result
4586 <result> = sub nsw <ty> <op1>, <op2> ; yields ty:result
4587 <result> = sub nuw nsw <ty> <op1>, <op2> ; yields ty:result
4592 The '``sub``' instruction returns the difference of its two operands.
4594 Note that the '``sub``' instruction is used to represent the '``neg``'
4595 instruction present in most other intermediate representations.
4600 The two arguments to the '``sub``' instruction must be
4601 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
4602 arguments must have identical types.
4607 The value produced is the integer difference of the two operands.
4609 If the difference has unsigned overflow, the result returned is the
4610 mathematical result modulo 2\ :sup:`n`\ , where n is the bit width of
4613 Because LLVM integers use a two's complement representation, this
4614 instruction is appropriate for both signed and unsigned integers.
4616 ``nuw`` and ``nsw`` stand for "No Unsigned Wrap" and "No Signed Wrap",
4617 respectively. If the ``nuw`` and/or ``nsw`` keywords are present, the
4618 result value of the ``sub`` is a :ref:`poison value <poisonvalues>` if
4619 unsigned and/or signed overflow, respectively, occurs.
4624 .. code-block:: llvm
4626 <result> = sub i32 4, %var ; yields i32:result = 4 - %var
4627 <result> = sub i32 0, %val ; yields i32:result = -%var
4631 '``fsub``' Instruction
4632 ^^^^^^^^^^^^^^^^^^^^^^
4639 <result> = fsub [fast-math flags]* <ty> <op1>, <op2> ; yields ty:result
4644 The '``fsub``' instruction returns the difference of its two operands.
4646 Note that the '``fsub``' instruction is used to represent the '``fneg``'
4647 instruction present in most other intermediate representations.
4652 The two arguments to the '``fsub``' instruction must be :ref:`floating
4653 point <t_floating>` or :ref:`vector <t_vector>` of floating point values.
4654 Both arguments must have identical types.
4659 The value produced is the floating point difference of the two operands.
4660 This instruction can also take any number of :ref:`fast-math
4661 flags <fastmath>`, which are optimization hints to enable otherwise
4662 unsafe floating point optimizations:
4667 .. code-block:: llvm
4669 <result> = fsub float 4.0, %var ; yields float:result = 4.0 - %var
4670 <result> = fsub float -0.0, %val ; yields float:result = -%var
4672 '``mul``' Instruction
4673 ^^^^^^^^^^^^^^^^^^^^^
4680 <result> = mul <ty> <op1>, <op2> ; yields ty:result
4681 <result> = mul nuw <ty> <op1>, <op2> ; yields ty:result
4682 <result> = mul nsw <ty> <op1>, <op2> ; yields ty:result
4683 <result> = mul nuw nsw <ty> <op1>, <op2> ; yields ty:result
4688 The '``mul``' instruction returns the product of its two operands.
4693 The two arguments to the '``mul``' instruction must be
4694 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
4695 arguments must have identical types.
4700 The value produced is the integer product of the two operands.
4702 If the result of the multiplication has unsigned overflow, the result
4703 returned is the mathematical result modulo 2\ :sup:`n`\ , where n is the
4704 bit width of the result.
4706 Because LLVM integers use a two's complement representation, and the
4707 result is the same width as the operands, this instruction returns the
4708 correct result for both signed and unsigned integers. If a full product
4709 (e.g. ``i32`` * ``i32`` -> ``i64``) is needed, the operands should be
4710 sign-extended or zero-extended as appropriate to the width of the full
4713 ``nuw`` and ``nsw`` stand for "No Unsigned Wrap" and "No Signed Wrap",
4714 respectively. If the ``nuw`` and/or ``nsw`` keywords are present, the
4715 result value of the ``mul`` is a :ref:`poison value <poisonvalues>` if
4716 unsigned and/or signed overflow, respectively, occurs.
4721 .. code-block:: llvm
4723 <result> = mul i32 4, %var ; yields i32:result = 4 * %var
4727 '``fmul``' Instruction
4728 ^^^^^^^^^^^^^^^^^^^^^^
4735 <result> = fmul [fast-math flags]* <ty> <op1>, <op2> ; yields ty:result
4740 The '``fmul``' instruction returns the product of its two operands.
4745 The two arguments to the '``fmul``' instruction must be :ref:`floating
4746 point <t_floating>` or :ref:`vector <t_vector>` of floating point values.
4747 Both arguments must have identical types.
4752 The value produced is the floating point product of the two operands.
4753 This instruction can also take any number of :ref:`fast-math
4754 flags <fastmath>`, which are optimization hints to enable otherwise
4755 unsafe floating point optimizations:
4760 .. code-block:: llvm
4762 <result> = fmul float 4.0, %var ; yields float:result = 4.0 * %var
4764 '``udiv``' Instruction
4765 ^^^^^^^^^^^^^^^^^^^^^^
4772 <result> = udiv <ty> <op1>, <op2> ; yields ty:result
4773 <result> = udiv exact <ty> <op1>, <op2> ; yields ty:result
4778 The '``udiv``' instruction returns the quotient of its two operands.
4783 The two arguments to the '``udiv``' instruction must be
4784 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
4785 arguments must have identical types.
4790 The value produced is the unsigned integer quotient of the two operands.
4792 Note that unsigned integer division and signed integer division are
4793 distinct operations; for signed integer division, use '``sdiv``'.
4795 Division by zero leads to undefined behavior.
4797 If the ``exact`` keyword is present, the result value of the ``udiv`` is
4798 a :ref:`poison value <poisonvalues>` if %op1 is not a multiple of %op2 (as
4799 such, "((a udiv exact b) mul b) == a").
4804 .. code-block:: llvm
4806 <result> = udiv i32 4, %var ; yields i32:result = 4 / %var
4808 '``sdiv``' Instruction
4809 ^^^^^^^^^^^^^^^^^^^^^^
4816 <result> = sdiv <ty> <op1>, <op2> ; yields ty:result
4817 <result> = sdiv exact <ty> <op1>, <op2> ; yields ty:result
4822 The '``sdiv``' instruction returns the quotient of its two operands.
4827 The two arguments to the '``sdiv``' instruction must be
4828 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
4829 arguments must have identical types.
4834 The value produced is the signed integer quotient of the two operands
4835 rounded towards zero.
4837 Note that signed integer division and unsigned integer division are
4838 distinct operations; for unsigned integer division, use '``udiv``'.
4840 Division by zero leads to undefined behavior. Overflow also leads to
4841 undefined behavior; this is a rare case, but can occur, for example, by
4842 doing a 32-bit division of -2147483648 by -1.
4844 If the ``exact`` keyword is present, the result value of the ``sdiv`` is
4845 a :ref:`poison value <poisonvalues>` if the result would be rounded.
4850 .. code-block:: llvm
4852 <result> = sdiv i32 4, %var ; yields i32:result = 4 / %var
4856 '``fdiv``' Instruction
4857 ^^^^^^^^^^^^^^^^^^^^^^
4864 <result> = fdiv [fast-math flags]* <ty> <op1>, <op2> ; yields ty:result
4869 The '``fdiv``' instruction returns the quotient of its two operands.
4874 The two arguments to the '``fdiv``' instruction must be :ref:`floating
4875 point <t_floating>` or :ref:`vector <t_vector>` of floating point values.
4876 Both arguments must have identical types.
4881 The value produced is the floating point quotient of the two operands.
4882 This instruction can also take any number of :ref:`fast-math
4883 flags <fastmath>`, which are optimization hints to enable otherwise
4884 unsafe floating point optimizations:
4889 .. code-block:: llvm
4891 <result> = fdiv float 4.0, %var ; yields float:result = 4.0 / %var
4893 '``urem``' Instruction
4894 ^^^^^^^^^^^^^^^^^^^^^^
4901 <result> = urem <ty> <op1>, <op2> ; yields ty:result
4906 The '``urem``' instruction returns the remainder from the unsigned
4907 division of its two arguments.
4912 The two arguments to the '``urem``' instruction must be
4913 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
4914 arguments must have identical types.
4919 This instruction returns the unsigned integer *remainder* of a division.
4920 This instruction always performs an unsigned division to get the
4923 Note that unsigned integer remainder and signed integer remainder are
4924 distinct operations; for signed integer remainder, use '``srem``'.
4926 Taking the remainder of a division by zero leads to undefined behavior.
4931 .. code-block:: llvm
4933 <result> = urem i32 4, %var ; yields i32:result = 4 % %var
4935 '``srem``' Instruction
4936 ^^^^^^^^^^^^^^^^^^^^^^
4943 <result> = srem <ty> <op1>, <op2> ; yields ty:result
4948 The '``srem``' instruction returns the remainder from the signed
4949 division of its two operands. This instruction can also take
4950 :ref:`vector <t_vector>` versions of the values in which case the elements
4956 The two arguments to the '``srem``' instruction must be
4957 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
4958 arguments must have identical types.
4963 This instruction returns the *remainder* of a division (where the result
4964 is either zero or has the same sign as the dividend, ``op1``), not the
4965 *modulo* operator (where the result is either zero or has the same sign
4966 as the divisor, ``op2``) of a value. For more information about the
4967 difference, see `The Math
4968 Forum <http://mathforum.org/dr.math/problems/anne.4.28.99.html>`_. For a
4969 table of how this is implemented in various languages, please see
4971 operation <http://en.wikipedia.org/wiki/Modulo_operation>`_.
4973 Note that signed integer remainder and unsigned integer remainder are
4974 distinct operations; for unsigned integer remainder, use '``urem``'.
4976 Taking the remainder of a division by zero leads to undefined behavior.
4977 Overflow also leads to undefined behavior; this is a rare case, but can
4978 occur, for example, by taking the remainder of a 32-bit division of
4979 -2147483648 by -1. (The remainder doesn't actually overflow, but this
4980 rule lets srem be implemented using instructions that return both the
4981 result of the division and the remainder.)
4986 .. code-block:: llvm
4988 <result> = srem i32 4, %var ; yields i32:result = 4 % %var
4992 '``frem``' Instruction
4993 ^^^^^^^^^^^^^^^^^^^^^^
5000 <result> = frem [fast-math flags]* <ty> <op1>, <op2> ; yields ty:result
5005 The '``frem``' instruction returns the remainder from the division of
5011 The two arguments to the '``frem``' instruction must be :ref:`floating
5012 point <t_floating>` or :ref:`vector <t_vector>` of floating point values.
5013 Both arguments must have identical types.
5018 This instruction returns the *remainder* of a division. The remainder
5019 has the same sign as the dividend. This instruction can also take any
5020 number of :ref:`fast-math flags <fastmath>`, which are optimization hints
5021 to enable otherwise unsafe floating point optimizations:
5026 .. code-block:: llvm
5028 <result> = frem float 4.0, %var ; yields float:result = 4.0 % %var
5032 Bitwise Binary Operations
5033 -------------------------
5035 Bitwise binary operators are used to do various forms of bit-twiddling
5036 in a program. They are generally very efficient instructions and can
5037 commonly be strength reduced from other instructions. They require two
5038 operands of the same type, execute an operation on them, and produce a
5039 single value. The resulting value is the same type as its operands.
5041 '``shl``' Instruction
5042 ^^^^^^^^^^^^^^^^^^^^^
5049 <result> = shl <ty> <op1>, <op2> ; yields ty:result
5050 <result> = shl nuw <ty> <op1>, <op2> ; yields ty:result
5051 <result> = shl nsw <ty> <op1>, <op2> ; yields ty:result
5052 <result> = shl nuw nsw <ty> <op1>, <op2> ; yields ty:result
5057 The '``shl``' instruction returns the first operand shifted to the left
5058 a specified number of bits.
5063 Both arguments to the '``shl``' instruction must be the same
5064 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer type.
5065 '``op2``' is treated as an unsigned value.
5070 The value produced is ``op1`` \* 2\ :sup:`op2` mod 2\ :sup:`n`,
5071 where ``n`` is the width of the result. If ``op2`` is (statically or
5072 dynamically) equal to or larger than the number of bits in
5073 ``op1``, the result is undefined. If the arguments are vectors, each
5074 vector element of ``op1`` is shifted by the corresponding shift amount
5077 If the ``nuw`` keyword is present, then the shift produces a :ref:`poison
5078 value <poisonvalues>` if it shifts out any non-zero bits. If the
5079 ``nsw`` keyword is present, then the shift produces a :ref:`poison
5080 value <poisonvalues>` if it shifts out any bits that disagree with the
5081 resultant sign bit. As such, NUW/NSW have the same semantics as they
5082 would if the shift were expressed as a mul instruction with the same
5083 nsw/nuw bits in (mul %op1, (shl 1, %op2)).
5088 .. code-block:: llvm
5090 <result> = shl i32 4, %var ; yields i32: 4 << %var
5091 <result> = shl i32 4, 2 ; yields i32: 16
5092 <result> = shl i32 1, 10 ; yields i32: 1024
5093 <result> = shl i32 1, 32 ; undefined
5094 <result> = shl <2 x i32> < i32 1, i32 1>, < i32 1, i32 2> ; yields: result=<2 x i32> < i32 2, i32 4>
5096 '``lshr``' Instruction
5097 ^^^^^^^^^^^^^^^^^^^^^^
5104 <result> = lshr <ty> <op1>, <op2> ; yields ty:result
5105 <result> = lshr exact <ty> <op1>, <op2> ; yields ty:result
5110 The '``lshr``' instruction (logical shift right) returns the first
5111 operand shifted to the right a specified number of bits with zero fill.
5116 Both arguments to the '``lshr``' instruction must be the same
5117 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer type.
5118 '``op2``' is treated as an unsigned value.
5123 This instruction always performs a logical shift right operation. The
5124 most significant bits of the result will be filled with zero bits after
5125 the shift. If ``op2`` is (statically or dynamically) equal to or larger
5126 than the number of bits in ``op1``, the result is undefined. If the
5127 arguments are vectors, each vector element of ``op1`` is shifted by the
5128 corresponding shift amount in ``op2``.
5130 If the ``exact`` keyword is present, the result value of the ``lshr`` is
5131 a :ref:`poison value <poisonvalues>` if any of the bits shifted out are
5137 .. code-block:: llvm
5139 <result> = lshr i32 4, 1 ; yields i32:result = 2
5140 <result> = lshr i32 4, 2 ; yields i32:result = 1
5141 <result> = lshr i8 4, 3 ; yields i8:result = 0
5142 <result> = lshr i8 -2, 1 ; yields i8:result = 0x7F
5143 <result> = lshr i32 1, 32 ; undefined
5144 <result> = lshr <2 x i32> < i32 -2, i32 4>, < i32 1, i32 2> ; yields: result=<2 x i32> < i32 0x7FFFFFFF, i32 1>
5146 '``ashr``' Instruction
5147 ^^^^^^^^^^^^^^^^^^^^^^
5154 <result> = ashr <ty> <op1>, <op2> ; yields ty:result
5155 <result> = ashr exact <ty> <op1>, <op2> ; yields ty:result
5160 The '``ashr``' instruction (arithmetic shift right) returns the first
5161 operand shifted to the right a specified number of bits with sign
5167 Both arguments to the '``ashr``' instruction must be the same
5168 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer type.
5169 '``op2``' is treated as an unsigned value.
5174 This instruction always performs an arithmetic shift right operation,
5175 The most significant bits of the result will be filled with the sign bit
5176 of ``op1``. If ``op2`` is (statically or dynamically) equal to or larger
5177 than the number of bits in ``op1``, the result is undefined. If the
5178 arguments are vectors, each vector element of ``op1`` is shifted by the
5179 corresponding shift amount in ``op2``.
5181 If the ``exact`` keyword is present, the result value of the ``ashr`` is
5182 a :ref:`poison value <poisonvalues>` if any of the bits shifted out are
5188 .. code-block:: llvm
5190 <result> = ashr i32 4, 1 ; yields i32:result = 2
5191 <result> = ashr i32 4, 2 ; yields i32:result = 1
5192 <result> = ashr i8 4, 3 ; yields i8:result = 0
5193 <result> = ashr i8 -2, 1 ; yields i8:result = -1
5194 <result> = ashr i32 1, 32 ; undefined
5195 <result> = ashr <2 x i32> < i32 -2, i32 4>, < i32 1, i32 3> ; yields: result=<2 x i32> < i32 -1, i32 0>
5197 '``and``' Instruction
5198 ^^^^^^^^^^^^^^^^^^^^^
5205 <result> = and <ty> <op1>, <op2> ; yields ty:result
5210 The '``and``' instruction returns the bitwise logical and of its two
5216 The two arguments to the '``and``' instruction must be
5217 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
5218 arguments must have identical types.
5223 The truth table used for the '``and``' instruction is:
5240 .. code-block:: llvm
5242 <result> = and i32 4, %var ; yields i32:result = 4 & %var
5243 <result> = and i32 15, 40 ; yields i32:result = 8
5244 <result> = and i32 4, 8 ; yields i32:result = 0
5246 '``or``' Instruction
5247 ^^^^^^^^^^^^^^^^^^^^
5254 <result> = or <ty> <op1>, <op2> ; yields ty:result
5259 The '``or``' instruction returns the bitwise logical inclusive or of its
5265 The two arguments to the '``or``' instruction must be
5266 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
5267 arguments must have identical types.
5272 The truth table used for the '``or``' instruction is:
5291 <result> = or i32 4, %var ; yields i32:result = 4 | %var
5292 <result> = or i32 15, 40 ; yields i32:result = 47
5293 <result> = or i32 4, 8 ; yields i32:result = 12
5295 '``xor``' Instruction
5296 ^^^^^^^^^^^^^^^^^^^^^
5303 <result> = xor <ty> <op1>, <op2> ; yields ty:result
5308 The '``xor``' instruction returns the bitwise logical exclusive or of
5309 its two operands. The ``xor`` is used to implement the "one's
5310 complement" operation, which is the "~" operator in C.
5315 The two arguments to the '``xor``' instruction must be
5316 :ref:`integer <t_integer>` or :ref:`vector <t_vector>` of integer values. Both
5317 arguments must have identical types.
5322 The truth table used for the '``xor``' instruction is:
5339 .. code-block:: llvm
5341 <result> = xor i32 4, %var ; yields i32:result = 4 ^ %var
5342 <result> = xor i32 15, 40 ; yields i32:result = 39
5343 <result> = xor i32 4, 8 ; yields i32:result = 12
5344 <result> = xor i32 %V, -1 ; yields i32:result = ~%V
5349 LLVM supports several instructions to represent vector operations in a
5350 target-independent manner. These instructions cover the element-access
5351 and vector-specific operations needed to process vectors effectively.
5352 While LLVM does directly support these vector operations, many
5353 sophisticated algorithms will want to use target-specific intrinsics to
5354 take full advantage of a specific target.
5356 .. _i_extractelement:
5358 '``extractelement``' Instruction
5359 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5366 <result> = extractelement <n x <ty>> <val>, <ty2> <idx> ; yields <ty>
5371 The '``extractelement``' instruction extracts a single scalar element
5372 from a vector at a specified index.
5377 The first operand of an '``extractelement``' instruction is a value of
5378 :ref:`vector <t_vector>` type. The second operand is an index indicating
5379 the position from which to extract the element. The index may be a
5380 variable of any integer type.
5385 The result is a scalar of the same type as the element type of ``val``.
5386 Its value is the value at position ``idx`` of ``val``. If ``idx``
5387 exceeds the length of ``val``, the results are undefined.
5392 .. code-block:: llvm
5394 <result> = extractelement <4 x i32> %vec, i32 0 ; yields i32
5396 .. _i_insertelement:
5398 '``insertelement``' Instruction
5399 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5406 <result> = insertelement <n x <ty>> <val>, <ty> <elt>, <ty2> <idx> ; yields <n x <ty>>
5411 The '``insertelement``' instruction inserts a scalar element into a
5412 vector at a specified index.
5417 The first operand of an '``insertelement``' instruction is a value of
5418 :ref:`vector <t_vector>` type. The second operand is a scalar value whose
5419 type must equal the element type of the first operand. The third operand
5420 is an index indicating the position at which to insert the value. The
5421 index may be a variable of any integer type.
5426 The result is a vector of the same type as ``val``. Its element values
5427 are those of ``val`` except at position ``idx``, where it gets the value
5428 ``elt``. If ``idx`` exceeds the length of ``val``, the results are
5434 .. code-block:: llvm
5436 <result> = insertelement <4 x i32> %vec, i32 1, i32 0 ; yields <4 x i32>
5438 .. _i_shufflevector:
5440 '``shufflevector``' Instruction
5441 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5448 <result> = shufflevector <n x <ty>> <v1>, <n x <ty>> <v2>, <m x i32> <mask> ; yields <m x <ty>>
5453 The '``shufflevector``' instruction constructs a permutation of elements
5454 from two input vectors, returning a vector with the same element type as
5455 the input and length that is the same as the shuffle mask.
5460 The first two operands of a '``shufflevector``' instruction are vectors
5461 with the same type. The third argument is a shuffle mask whose element
5462 type is always 'i32'. The result of the instruction is a vector whose
5463 length is the same as the shuffle mask and whose element type is the
5464 same as the element type of the first two operands.
5466 The shuffle mask operand is required to be a constant vector with either
5467 constant integer or undef values.
5472 The elements of the two input vectors are numbered from left to right
5473 across both of the vectors. The shuffle mask operand specifies, for each
5474 element of the result vector, which element of the two input vectors the
5475 result element gets. The element selector may be undef (meaning "don't
5476 care") and the second operand may be undef if performing a shuffle from
5482 .. code-block:: llvm
5484 <result> = shufflevector <4 x i32> %v1, <4 x i32> %v2,
5485 <4 x i32> <i32 0, i32 4, i32 1, i32 5> ; yields <4 x i32>
5486 <result> = shufflevector <4 x i32> %v1, <4 x i32> undef,
5487 <4 x i32> <i32 0, i32 1, i32 2, i32 3> ; yields <4 x i32> - Identity shuffle.
5488 <result> = shufflevector <8 x i32> %v1, <8 x i32> undef,
5489 <4 x i32> <i32 0, i32 1, i32 2, i32 3> ; yields <4 x i32>
5490 <result> = shufflevector <4 x i32> %v1, <4 x i32> %v2,
5491 <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7 > ; yields <8 x i32>
5493 Aggregate Operations
5494 --------------------
5496 LLVM supports several instructions for working with
5497 :ref:`aggregate <t_aggregate>` values.
5501 '``extractvalue``' Instruction
5502 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5509 <result> = extractvalue <aggregate type> <val>, <idx>{, <idx>}*
5514 The '``extractvalue``' instruction extracts the value of a member field
5515 from an :ref:`aggregate <t_aggregate>` value.
5520 The first operand of an '``extractvalue``' instruction is a value of
5521 :ref:`struct <t_struct>` or :ref:`array <t_array>` type. The operands are
5522 constant indices to specify which value to extract in a similar manner
5523 as indices in a '``getelementptr``' instruction.
5525 The major differences to ``getelementptr`` indexing are:
5527 - Since the value being indexed is not a pointer, the first index is
5528 omitted and assumed to be zero.
5529 - At least one index must be specified.
5530 - Not only struct indices but also array indices must be in bounds.
5535 The result is the value at the position in the aggregate specified by
5541 .. code-block:: llvm
5543 <result> = extractvalue {i32, float} %agg, 0 ; yields i32
5547 '``insertvalue``' Instruction
5548 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
5555 <result> = insertvalue <aggregate type> <val>, <ty> <elt>, <idx>{, <idx>}* ; yields <aggregate type>
5560 The '``insertvalue``' instruction inserts a value into a member field in
5561 an :ref:`aggregate <t_aggregate>` value.
5566 The first operand of an '``insertvalue``' instruction is a value of
5567 :ref:`struct <t_struct>` or :ref:`array <t_array>` type. The second operand is
5568 a first-class value to insert. The following operands are constant
5569 indices indicating the position at which to insert the value in a
5570 similar manner as indices in a '``extractvalue``' instruction. The value
5571 to insert must have the same type as the value identified by the
5577 The result is an aggregate of the same type as ``val``. Its value is
5578 that of ``val`` except that the value at the position specified by the
5579 indices is that of ``elt``.
5584 .. code-block:: llvm
5586 %agg1 = insertvalue {i32, float} undef, i32 1, 0 ; yields {i32 1, float undef}
5587 %agg2 = insertvalue {i32, float} %agg1, float %val, 1 ; yields {i32 1, float %val}
5588 %agg3 = insertvalue {i32, {float}} undef, float %val, 1, 0 ; yields {i32 undef, {float %val}}
5592 Memory Access and Addressing Operations
5593 ---------------------------------------
5595 A key design point of an SSA-based representation is how it represents
5596 memory. In LLVM, no memory locations are in SSA form, which makes things
5597 very simple. This section describes how to read, write, and allocate
5602 '``alloca``' Instruction
5603 ^^^^^^^^^^^^^^^^^^^^^^^^
5610 <result> = alloca [inalloca] <type> [, <ty> <NumElements>] [, align <alignment>] ; yields type*:result
5615 The '``alloca``' instruction allocates memory on the stack frame of the
5616 currently executing function, to be automatically released when this
5617 function returns to its caller. The object is always allocated in the
5618 generic address space (address space zero).
5623 The '``alloca``' instruction allocates ``sizeof(<type>)*NumElements``
5624 bytes of memory on the runtime stack, returning a pointer of the
5625 appropriate type to the program. If "NumElements" is specified, it is
5626 the number of elements allocated, otherwise "NumElements" is defaulted
5627 to be one. If a constant alignment is specified, the value result of the
5628 allocation is guaranteed to be aligned to at least that boundary. The
5629 alignment may not be greater than ``1 << 29``. If not specified, or if
5630 zero, the target can choose to align the allocation on any convenient
5631 boundary compatible with the type.
5633 '``type``' may be any sized type.
5638 Memory is allocated; a pointer is returned. The operation is undefined
5639 if there is insufficient stack space for the allocation. '``alloca``'d
5640 memory is automatically released when the function returns. The
5641 '``alloca``' instruction is commonly used to represent automatic
5642 variables that must have an address available. When the function returns
5643 (either with the ``ret`` or ``resume`` instructions), the memory is
5644 reclaimed. Allocating zero bytes is legal, but the result is undefined.
5645 The order in which memory is allocated (ie., which way the stack grows)
5651 .. code-block:: llvm
5653 %ptr = alloca i32 ; yields i32*:ptr
5654 %ptr = alloca i32, i32 4 ; yields i32*:ptr
5655 %ptr = alloca i32, i32 4, align 1024 ; yields i32*:ptr
5656 %ptr = alloca i32, align 1024 ; yields i32*:ptr
5660 '``load``' Instruction
5661 ^^^^^^^^^^^^^^^^^^^^^^
5668 <result> = load [volatile] <ty>, <ty>* <pointer>[, align <alignment>][, !nontemporal !<index>][, !invariant.load !<index>][, !nonnull !<index>][, !dereferenceable !<index>][, !dereferenceable_or_null !<index>]
5669 <result> = load atomic [volatile] <ty>* <pointer> [singlethread] <ordering>, align <alignment>
5670 !<index> = !{ i32 1 }
5675 The '``load``' instruction is used to read from memory.
5680 The argument to the ``load`` instruction specifies the memory address
5681 from which to load. The type specified must be a :ref:`first
5682 class <t_firstclass>` type. If the ``load`` is marked as ``volatile``,
5683 then the optimizer is not allowed to modify the number or order of
5684 execution of this ``load`` with other :ref:`volatile
5685 operations <volatile>`.
5687 If the ``load`` is marked as ``atomic``, it takes an extra
5688 :ref:`ordering <ordering>` and optional ``singlethread`` argument. The
5689 ``release`` and ``acq_rel`` orderings are not valid on ``load``
5690 instructions. Atomic loads produce :ref:`defined <memmodel>` results
5691 when they may see multiple atomic stores. The type of the pointee must
5692 be an integer type whose bit width is a power of two greater than or
5693 equal to eight and less than or equal to a target-specific size limit.
5694 ``align`` must be explicitly specified on atomic loads, and the load has
5695 undefined behavior if the alignment is not set to a value which is at
5696 least the size in bytes of the pointee. ``!nontemporal`` does not have
5697 any defined semantics for atomic loads.
5699 The optional constant ``align`` argument specifies the alignment of the
5700 operation (that is, the alignment of the memory address). A value of 0
5701 or an omitted ``align`` argument means that the operation has the ABI
5702 alignment for the target. It is the responsibility of the code emitter
5703 to ensure that the alignment information is correct. Overestimating the
5704 alignment results in undefined behavior. Underestimating the alignment
5705 may produce less efficient code. An alignment of 1 is always safe. The
5706 maximum possible alignment is ``1 << 29``.
5708 The optional ``!nontemporal`` metadata must reference a single
5709 metadata name ``<index>`` corresponding to a metadata node with one
5710 ``i32`` entry of value 1. The existence of the ``!nontemporal``
5711 metadata on the instruction tells the optimizer and code generator
5712 that this load is not expected to be reused in the cache. The code
5713 generator may select special instructions to save cache bandwidth, such
5714 as the ``MOVNT`` instruction on x86.
5716 The optional ``!invariant.load`` metadata must reference a single
5717 metadata name ``<index>`` corresponding to a metadata node with no
5718 entries. The existence of the ``!invariant.load`` metadata on the
5719 instruction tells the optimizer and code generator that the address
5720 operand to this load points to memory which can be assumed unchanged.
5721 Being invariant does not imply that a location is dereferenceable,
5722 but it does imply that once the location is known dereferenceable
5723 its value is henceforth unchanging.
5725 The optional ``!nonnull`` metadata must reference a single
5726 metadata name ``<index>`` corresponding to a metadata node with no
5727 entries. The existence of the ``!nonnull`` metadata on the
5728 instruction tells the optimizer that the value loaded is known to
5729 never be null. This is analogous to the ''nonnull'' attribute
5730 on parameters and return values. This metadata can only be applied
5731 to loads of a pointer type.
5733 The optional ``!dereferenceable`` metadata must reference a single
5734 metadata name ``<index>`` corresponding to a metadata node with one ``i64``
5735 entry. The existence of the ``!dereferenceable`` metadata on the instruction
5736 tells the optimizer that the value loaded is known to be dereferenceable.
5737 The number of bytes known to be dereferenceable is specified by the integer
5738 value in the metadata node. This is analogous to the ''dereferenceable''
5739 attribute on parameters and return values. This metadata can only be applied
5740 to loads of a pointer type.
5742 The optional ``!dereferenceable_or_null`` metadata must reference a single
5743 metadata name ``<index>`` corresponding to a metadata node with one ``i64``
5744 entry. The existence of the ``!dereferenceable_or_null`` metadata on the
5745 instruction tells the optimizer that the value loaded is known to be either
5746 dereferenceable or null.
5747 The number of bytes known to be dereferenceable is specified by the integer
5748 value in the metadata node. This is analogous to the ''dereferenceable_or_null''
5749 attribute on parameters and return values. This metadata can only be applied
5750 to loads of a pointer type.
5755 The location of memory pointed to is loaded. If the value being loaded
5756 is of scalar type then the number of bytes read does not exceed the
5757 minimum number of bytes needed to hold all bits of the type. For
5758 example, loading an ``i24`` reads at most three bytes. When loading a
5759 value of a type like ``i20`` with a size that is not an integral number
5760 of bytes, the result is undefined if the value was not originally
5761 written using a store of the same type.
5766 .. code-block:: llvm
5768 %ptr = alloca i32 ; yields i32*:ptr
5769 store i32 3, i32* %ptr ; yields void
5770 %val = load i32, i32* %ptr ; yields i32:val = i32 3
5774 '``store``' Instruction
5775 ^^^^^^^^^^^^^^^^^^^^^^^
5782 store [volatile] <ty> <value>, <ty>* <pointer>[, align <alignment>][, !nontemporal !<index>] ; yields void
5783 store atomic [volatile] <ty> <value>, <ty>* <pointer> [singlethread] <ordering>, align <alignment> ; yields void
5788 The '``store``' instruction is used to write to memory.
5793 There are two arguments to the ``store`` instruction: a value to store
5794 and an address at which to store it. The type of the ``<pointer>``
5795 operand must be a pointer to the :ref:`first class <t_firstclass>` type of
5796 the ``<value>`` operand. If the ``store`` is marked as ``volatile``,
5797 then the optimizer is not allowed to modify the number or order of
5798 execution of this ``store`` with other :ref:`volatile
5799 operations <volatile>`.
5801 If the ``store`` is marked as ``atomic``, it takes an extra
5802 :ref:`ordering <ordering>` and optional ``singlethread`` argument. The
5803 ``acquire`` and ``acq_rel`` orderings aren't valid on ``store``
5804 instructions. Atomic loads produce :ref:`defined <memmodel>` results
5805 when they may see multiple atomic stores. The type of the pointee must
5806 be an integer type whose bit width is a power of two greater than or
5807 equal to eight and less than or equal to a target-specific size limit.
5808 ``align`` must be explicitly specified on atomic stores, and the store
5809 has undefined behavior if the alignment is not set to a value which is
5810 at least the size in bytes of the pointee. ``!nontemporal`` does not
5811 have any defined semantics for atomic stores.
5813 The optional constant ``align`` argument specifies the alignment of the
5814 operation (that is, the alignment of the memory address). A value of 0
5815 or an omitted ``align`` argument means that the operation has the ABI
5816 alignment for the target. It is the responsibility of the code emitter
5817 to ensure that the alignment information is correct. Overestimating the
5818 alignment results in undefined behavior. Underestimating the
5819 alignment may produce less efficient code. An alignment of 1 is always
5820 safe. The maximum possible alignment is ``1 << 29``.
5822 The optional ``!nontemporal`` metadata must reference a single metadata
5823 name ``<index>`` corresponding to a metadata node with one ``i32`` entry of
5824 value 1. The existence of the ``!nontemporal`` metadata on the instruction
5825 tells the optimizer and code generator that this load is not expected to
5826 be reused in the cache. The code generator may select special
5827 instructions to save cache bandwidth, such as the MOVNT instruction on
5833 The contents of memory are updated to contain ``<value>`` at the
5834 location specified by the ``<pointer>`` operand. If ``<value>`` is
5835 of scalar type then the number of bytes written does not exceed the
5836 minimum number of bytes needed to hold all bits of the type. For
5837 example, storing an ``i24`` writes at most three bytes. When writing a
5838 value of a type like ``i20`` with a size that is not an integral number
5839 of bytes, it is unspecified what happens to the extra bits that do not
5840 belong to the type, but they will typically be overwritten.
5845 .. code-block:: llvm
5847 %ptr = alloca i32 ; yields i32*:ptr
5848 store i32 3, i32* %ptr ; yields void
5849 %val = load i32* %ptr ; yields i32:val = i32 3
5853 '``fence``' Instruction
5854 ^^^^^^^^^^^^^^^^^^^^^^^
5861 fence [singlethread] <ordering> ; yields void
5866 The '``fence``' instruction is used to introduce happens-before edges
5872 '``fence``' instructions take an :ref:`ordering <ordering>` argument which
5873 defines what *synchronizes-with* edges they add. They can only be given
5874 ``acquire``, ``release``, ``acq_rel``, and ``seq_cst`` orderings.
5879 A fence A which has (at least) ``release`` ordering semantics
5880 *synchronizes with* a fence B with (at least) ``acquire`` ordering
5881 semantics if and only if there exist atomic operations X and Y, both
5882 operating on some atomic object M, such that A is sequenced before X, X
5883 modifies M (either directly or through some side effect of a sequence
5884 headed by X), Y is sequenced before B, and Y observes M. This provides a
5885 *happens-before* dependency between A and B. Rather than an explicit
5886 ``fence``, one (but not both) of the atomic operations X or Y might
5887 provide a ``release`` or ``acquire`` (resp.) ordering constraint and
5888 still *synchronize-with* the explicit ``fence`` and establish the
5889 *happens-before* edge.
5891 A ``fence`` which has ``seq_cst`` ordering, in addition to having both
5892 ``acquire`` and ``release`` semantics specified above, participates in
5893 the global program order of other ``seq_cst`` operations and/or fences.
5895 The optional ":ref:`singlethread <singlethread>`" argument specifies
5896 that the fence only synchronizes with other fences in the same thread.
5897 (This is useful for interacting with signal handlers.)
5902 .. code-block:: llvm
5904 fence acquire ; yields void
5905 fence singlethread seq_cst ; yields void
5909 '``cmpxchg``' Instruction
5910 ^^^^^^^^^^^^^^^^^^^^^^^^^
5917 cmpxchg [weak] [volatile] <ty>* <pointer>, <ty> <cmp>, <ty> <new> [singlethread] <success ordering> <failure ordering> ; yields { ty, i1 }
5922 The '``cmpxchg``' instruction is used to atomically modify memory. It
5923 loads a value in memory and compares it to a given value. If they are
5924 equal, it tries to store a new value into the memory.
5929 There are three arguments to the '``cmpxchg``' instruction: an address
5930 to operate on, a value to compare to the value currently be at that
5931 address, and a new value to place at that address if the compared values
5932 are equal. The type of '<cmp>' must be an integer type whose bit width
5933 is a power of two greater than or equal to eight and less than or equal
5934 to a target-specific size limit. '<cmp>' and '<new>' must have the same
5935 type, and the type of '<pointer>' must be a pointer to that type. If the
5936 ``cmpxchg`` is marked as ``volatile``, then the optimizer is not allowed
5937 to modify the number or order of execution of this ``cmpxchg`` with
5938 other :ref:`volatile operations <volatile>`.
5940 The success and failure :ref:`ordering <ordering>` arguments specify how this
5941 ``cmpxchg`` synchronizes with other atomic operations. Both ordering parameters
5942 must be at least ``monotonic``, the ordering constraint on failure must be no
5943 stronger than that on success, and the failure ordering cannot be either
5944 ``release`` or ``acq_rel``.
5946 The optional "``singlethread``" argument declares that the ``cmpxchg``
5947 is only atomic with respect to code (usually signal handlers) running in
5948 the same thread as the ``cmpxchg``. Otherwise the cmpxchg is atomic with
5949 respect to all other code in the system.
5951 The pointer passed into cmpxchg must have alignment greater than or
5952 equal to the size in memory of the operand.
5957 The contents of memory at the location specified by the '``<pointer>``' operand
5958 is read and compared to '``<cmp>``'; if the read value is the equal, the
5959 '``<new>``' is written. The original value at the location is returned, together
5960 with a flag indicating success (true) or failure (false).
5962 If the cmpxchg operation is marked as ``weak`` then a spurious failure is
5963 permitted: the operation may not write ``<new>`` even if the comparison
5966 If the cmpxchg operation is strong (the default), the i1 value is 1 if and only
5967 if the value loaded equals ``cmp``.
5969 A successful ``cmpxchg`` is a read-modify-write instruction for the purpose of
5970 identifying release sequences. A failed ``cmpxchg`` is equivalent to an atomic
5971 load with an ordering parameter determined the second ordering parameter.
5976 .. code-block:: llvm
5979 %orig = atomic load i32, i32* %ptr unordered ; yields i32
5983 %cmp = phi i32 [ %orig, %entry ], [%old, %loop]
5984 %squared = mul i32 %cmp, %cmp
5985 %val_success = cmpxchg i32* %ptr, i32 %cmp, i32 %squared acq_rel monotonic ; yields { i32, i1 }
5986 %value_loaded = extractvalue { i32, i1 } %val_success, 0
5987 %success = extractvalue { i32, i1 } %val_success, 1
5988 br i1 %success, label %done, label %loop
5995 '``atomicrmw``' Instruction
5996 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
6003 atomicrmw [volatile] <operation> <ty>* <pointer>, <ty> <value> [singlethread] <ordering> ; yields ty
6008 The '``atomicrmw``' instruction is used to atomically modify memory.
6013 There are three arguments to the '``atomicrmw``' instruction: an
6014 operation to apply, an address whose value to modify, an argument to the
6015 operation. The operation must be one of the following keywords:
6029 The type of '<value>' must be an integer type whose bit width is a power
6030 of two greater than or equal to eight and less than or equal to a
6031 target-specific size limit. The type of the '``<pointer>``' operand must
6032 be a pointer to that type. If the ``atomicrmw`` is marked as
6033 ``volatile``, then the optimizer is not allowed to modify the number or
6034 order of execution of this ``atomicrmw`` with other :ref:`volatile
6035 operations <volatile>`.
6040 The contents of memory at the location specified by the '``<pointer>``'
6041 operand are atomically read, modified, and written back. The original
6042 value at the location is returned. The modification is specified by the
6045 - xchg: ``*ptr = val``
6046 - add: ``*ptr = *ptr + val``
6047 - sub: ``*ptr = *ptr - val``
6048 - and: ``*ptr = *ptr & val``
6049 - nand: ``*ptr = ~(*ptr & val)``
6050 - or: ``*ptr = *ptr | val``
6051 - xor: ``*ptr = *ptr ^ val``
6052 - max: ``*ptr = *ptr > val ? *ptr : val`` (using a signed comparison)
6053 - min: ``*ptr = *ptr < val ? *ptr : val`` (using a signed comparison)
6054 - umax: ``*ptr = *ptr > val ? *ptr : val`` (using an unsigned
6056 - umin: ``*ptr = *ptr < val ? *ptr : val`` (using an unsigned
6062 .. code-block:: llvm
6064 %old = atomicrmw add i32* %ptr, i32 1 acquire ; yields i32
6066 .. _i_getelementptr:
6068 '``getelementptr``' Instruction
6069 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6076 <result> = getelementptr <ty>, <ty>* <ptrval>{, <ty> <idx>}*
6077 <result> = getelementptr inbounds <ty>, <ty>* <ptrval>{, <ty> <idx>}*
6078 <result> = getelementptr <ty>, <ptr vector> <ptrval>, <vector index type> <idx>
6083 The '``getelementptr``' instruction is used to get the address of a
6084 subelement of an :ref:`aggregate <t_aggregate>` data structure. It performs
6085 address calculation only and does not access memory.
6090 The first argument is always a type used as the basis for the calculations.
6091 The second argument is always a pointer or a vector of pointers, and is the
6092 base address to start from. The remaining arguments are indices
6093 that indicate which of the elements of the aggregate object are indexed.
6094 The interpretation of each index is dependent on the type being indexed
6095 into. The first index always indexes the pointer value given as the
6096 first argument, the second index indexes a value of the type pointed to
6097 (not necessarily the value directly pointed to, since the first index
6098 can be non-zero), etc. The first type indexed into must be a pointer
6099 value, subsequent types can be arrays, vectors, and structs. Note that
6100 subsequent types being indexed into can never be pointers, since that
6101 would require loading the pointer before continuing calculation.
6103 The type of each index argument depends on the type it is indexing into.
6104 When indexing into a (optionally packed) structure, only ``i32`` integer
6105 **constants** are allowed (when using a vector of indices they must all
6106 be the **same** ``i32`` integer constant). When indexing into an array,
6107 pointer or vector, integers of any width are allowed, and they are not
6108 required to be constant. These integers are treated as signed values
6111 For example, let's consider a C code fragment and how it gets compiled
6127 int *foo(struct ST *s) {
6128 return &s[1].Z.B[5][13];
6131 The LLVM code generated by Clang is:
6133 .. code-block:: llvm
6135 %struct.RT = type { i8, [10 x [20 x i32]], i8 }
6136 %struct.ST = type { i32, double, %struct.RT }
6138 define i32* @foo(%struct.ST* %s) nounwind uwtable readnone optsize ssp {
6140 %arrayidx = getelementptr inbounds %struct.ST, %struct.ST* %s, i64 1, i32 2, i32 1, i64 5, i64 13
6147 In the example above, the first index is indexing into the
6148 '``%struct.ST*``' type, which is a pointer, yielding a '``%struct.ST``'
6149 = '``{ i32, double, %struct.RT }``' type, a structure. The second index
6150 indexes into the third element of the structure, yielding a
6151 '``%struct.RT``' = '``{ i8 , [10 x [20 x i32]], i8 }``' type, another
6152 structure. The third index indexes into the second element of the
6153 structure, yielding a '``[10 x [20 x i32]]``' type, an array. The two
6154 dimensions of the array are subscripted into, yielding an '``i32``'
6155 type. The '``getelementptr``' instruction returns a pointer to this
6156 element, thus computing a value of '``i32*``' type.
6158 Note that it is perfectly legal to index partially through a structure,
6159 returning a pointer to an inner element. Because of this, the LLVM code
6160 for the given testcase is equivalent to:
6162 .. code-block:: llvm
6164 define i32* @foo(%struct.ST* %s) {
6165 %t1 = getelementptr %struct.ST, %struct.ST* %s, i32 1 ; yields %struct.ST*:%t1
6166 %t2 = getelementptr %struct.ST, %struct.ST* %t1, i32 0, i32 2 ; yields %struct.RT*:%t2
6167 %t3 = getelementptr %struct.RT, %struct.RT* %t2, i32 0, i32 1 ; yields [10 x [20 x i32]]*:%t3
6168 %t4 = getelementptr [10 x [20 x i32]], [10 x [20 x i32]]* %t3, i32 0, i32 5 ; yields [20 x i32]*:%t4
6169 %t5 = getelementptr [20 x i32], [20 x i32]* %t4, i32 0, i32 13 ; yields i32*:%t5
6173 If the ``inbounds`` keyword is present, the result value of the
6174 ``getelementptr`` is a :ref:`poison value <poisonvalues>` if the base
6175 pointer is not an *in bounds* address of an allocated object, or if any
6176 of the addresses that would be formed by successive addition of the
6177 offsets implied by the indices to the base address with infinitely
6178 precise signed arithmetic are not an *in bounds* address of that
6179 allocated object. The *in bounds* addresses for an allocated object are
6180 all the addresses that point into the object, plus the address one byte
6181 past the end. In cases where the base is a vector of pointers the
6182 ``inbounds`` keyword applies to each of the computations element-wise.
6184 If the ``inbounds`` keyword is not present, the offsets are added to the
6185 base address with silently-wrapping two's complement arithmetic. If the
6186 offsets have a different width from the pointer, they are sign-extended
6187 or truncated to the width of the pointer. The result value of the
6188 ``getelementptr`` may be outside the object pointed to by the base
6189 pointer. The result value may not necessarily be used to access memory
6190 though, even if it happens to point into allocated storage. See the
6191 :ref:`Pointer Aliasing Rules <pointeraliasing>` section for more
6194 The getelementptr instruction is often confusing. For some more insight
6195 into how it works, see :doc:`the getelementptr FAQ <GetElementPtr>`.
6200 .. code-block:: llvm
6202 ; yields [12 x i8]*:aptr
6203 %aptr = getelementptr {i32, [12 x i8]}, {i32, [12 x i8]}* %saptr, i64 0, i32 1
6205 %vptr = getelementptr {i32, <2 x i8>}, {i32, <2 x i8>}* %svptr, i64 0, i32 1, i32 1
6207 %eptr = getelementptr [12 x i8], [12 x i8]* %aptr, i64 0, i32 1
6209 %iptr = getelementptr [10 x i32], [10 x i32]* @arr, i16 0, i16 0
6211 In cases where the pointer argument is a vector of pointers, each index
6212 must be a vector with the same number of elements. For example:
6214 .. code-block:: llvm
6216 %A = getelementptr i8, <4 x i8*> %ptrs, <4 x i64> %offsets,
6218 Conversion Operations
6219 ---------------------
6221 The instructions in this category are the conversion instructions
6222 (casting) which all take a single operand and a type. They perform
6223 various bit conversions on the operand.
6225 '``trunc .. to``' Instruction
6226 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6233 <result> = trunc <ty> <value> to <ty2> ; yields ty2
6238 The '``trunc``' instruction truncates its operand to the type ``ty2``.
6243 The '``trunc``' instruction takes a value to trunc, and a type to trunc
6244 it to. Both types must be of :ref:`integer <t_integer>` types, or vectors
6245 of the same number of integers. The bit size of the ``value`` must be
6246 larger than the bit size of the destination type, ``ty2``. Equal sized
6247 types are not allowed.
6252 The '``trunc``' instruction truncates the high order bits in ``value``
6253 and converts the remaining bits to ``ty2``. Since the source size must
6254 be larger than the destination size, ``trunc`` cannot be a *no-op cast*.
6255 It will always truncate bits.
6260 .. code-block:: llvm
6262 %X = trunc i32 257 to i8 ; yields i8:1
6263 %Y = trunc i32 123 to i1 ; yields i1:true
6264 %Z = trunc i32 122 to i1 ; yields i1:false
6265 %W = trunc <2 x i16> <i16 8, i16 7> to <2 x i8> ; yields <i8 8, i8 7>
6267 '``zext .. to``' Instruction
6268 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6275 <result> = zext <ty> <value> to <ty2> ; yields ty2
6280 The '``zext``' instruction zero extends its operand to type ``ty2``.
6285 The '``zext``' instruction takes a value to cast, and a type to cast it
6286 to. Both types must be of :ref:`integer <t_integer>` types, or vectors of
6287 the same number of integers. The bit size of the ``value`` must be
6288 smaller than the bit size of the destination type, ``ty2``.
6293 The ``zext`` fills the high order bits of the ``value`` with zero bits
6294 until it reaches the size of the destination type, ``ty2``.
6296 When zero extending from i1, the result will always be either 0 or 1.
6301 .. code-block:: llvm
6303 %X = zext i32 257 to i64 ; yields i64:257
6304 %Y = zext i1 true to i32 ; yields i32:1
6305 %Z = zext <2 x i16> <i16 8, i16 7> to <2 x i32> ; yields <i32 8, i32 7>
6307 '``sext .. to``' Instruction
6308 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6315 <result> = sext <ty> <value> to <ty2> ; yields ty2
6320 The '``sext``' sign extends ``value`` to the type ``ty2``.
6325 The '``sext``' instruction takes a value to cast, and a type to cast it
6326 to. Both types must be of :ref:`integer <t_integer>` types, or vectors of
6327 the same number of integers. The bit size of the ``value`` must be
6328 smaller than the bit size of the destination type, ``ty2``.
6333 The '``sext``' instruction performs a sign extension by copying the sign
6334 bit (highest order bit) of the ``value`` until it reaches the bit size
6335 of the type ``ty2``.
6337 When sign extending from i1, the extension always results in -1 or 0.
6342 .. code-block:: llvm
6344 %X = sext i8 -1 to i16 ; yields i16 :65535
6345 %Y = sext i1 true to i32 ; yields i32:-1
6346 %Z = sext <2 x i16> <i16 8, i16 7> to <2 x i32> ; yields <i32 8, i32 7>
6348 '``fptrunc .. to``' Instruction
6349 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6356 <result> = fptrunc <ty> <value> to <ty2> ; yields ty2
6361 The '``fptrunc``' instruction truncates ``value`` to type ``ty2``.
6366 The '``fptrunc``' instruction takes a :ref:`floating point <t_floating>`
6367 value to cast and a :ref:`floating point <t_floating>` type to cast it to.
6368 The size of ``value`` must be larger than the size of ``ty2``. This
6369 implies that ``fptrunc`` cannot be used to make a *no-op cast*.
6374 The '``fptrunc``' instruction truncates a ``value`` from a larger
6375 :ref:`floating point <t_floating>` type to a smaller :ref:`floating
6376 point <t_floating>` type. If the value cannot fit within the
6377 destination type, ``ty2``, then the results are undefined.
6382 .. code-block:: llvm
6384 %X = fptrunc double 123.0 to float ; yields float:123.0
6385 %Y = fptrunc double 1.0E+300 to float ; yields undefined
6387 '``fpext .. to``' Instruction
6388 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6395 <result> = fpext <ty> <value> to <ty2> ; yields ty2
6400 The '``fpext``' extends a floating point ``value`` to a larger floating
6406 The '``fpext``' instruction takes a :ref:`floating point <t_floating>`
6407 ``value`` to cast, and a :ref:`floating point <t_floating>` type to cast it
6408 to. The source type must be smaller than the destination type.
6413 The '``fpext``' instruction extends the ``value`` from a smaller
6414 :ref:`floating point <t_floating>` type to a larger :ref:`floating
6415 point <t_floating>` type. The ``fpext`` cannot be used to make a
6416 *no-op cast* because it always changes bits. Use ``bitcast`` to make a
6417 *no-op cast* for a floating point cast.
6422 .. code-block:: llvm
6424 %X = fpext float 3.125 to double ; yields double:3.125000e+00
6425 %Y = fpext double %X to fp128 ; yields fp128:0xL00000000000000004000900000000000
6427 '``fptoui .. to``' Instruction
6428 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6435 <result> = fptoui <ty> <value> to <ty2> ; yields ty2
6440 The '``fptoui``' converts a floating point ``value`` to its unsigned
6441 integer equivalent of type ``ty2``.
6446 The '``fptoui``' instruction takes a value to cast, which must be a
6447 scalar or vector :ref:`floating point <t_floating>` value, and a type to
6448 cast it to ``ty2``, which must be an :ref:`integer <t_integer>` type. If
6449 ``ty`` is a vector floating point type, ``ty2`` must be a vector integer
6450 type with the same number of elements as ``ty``
6455 The '``fptoui``' instruction converts its :ref:`floating
6456 point <t_floating>` operand into the nearest (rounding towards zero)
6457 unsigned integer value. If the value cannot fit in ``ty2``, the results
6463 .. code-block:: llvm
6465 %X = fptoui double 123.0 to i32 ; yields i32:123
6466 %Y = fptoui float 1.0E+300 to i1 ; yields undefined:1
6467 %Z = fptoui float 1.04E+17 to i8 ; yields undefined:1
6469 '``fptosi .. to``' Instruction
6470 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6477 <result> = fptosi <ty> <value> to <ty2> ; yields ty2
6482 The '``fptosi``' instruction converts :ref:`floating point <t_floating>`
6483 ``value`` to type ``ty2``.
6488 The '``fptosi``' instruction takes a value to cast, which must be a
6489 scalar or vector :ref:`floating point <t_floating>` value, and a type to
6490 cast it to ``ty2``, which must be an :ref:`integer <t_integer>` type. If
6491 ``ty`` is a vector floating point type, ``ty2`` must be a vector integer
6492 type with the same number of elements as ``ty``
6497 The '``fptosi``' instruction converts its :ref:`floating
6498 point <t_floating>` operand into the nearest (rounding towards zero)
6499 signed integer value. If the value cannot fit in ``ty2``, the results
6505 .. code-block:: llvm
6507 %X = fptosi double -123.0 to i32 ; yields i32:-123
6508 %Y = fptosi float 1.0E-247 to i1 ; yields undefined:1
6509 %Z = fptosi float 1.04E+17 to i8 ; yields undefined:1
6511 '``uitofp .. to``' Instruction
6512 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6519 <result> = uitofp <ty> <value> to <ty2> ; yields ty2
6524 The '``uitofp``' instruction regards ``value`` as an unsigned integer
6525 and converts that value to the ``ty2`` type.
6530 The '``uitofp``' instruction takes a value to cast, which must be a
6531 scalar or vector :ref:`integer <t_integer>` value, and a type to cast it to
6532 ``ty2``, which must be an :ref:`floating point <t_floating>` type. If
6533 ``ty`` is a vector integer type, ``ty2`` must be a vector floating point
6534 type with the same number of elements as ``ty``
6539 The '``uitofp``' instruction interprets its operand as an unsigned
6540 integer quantity and converts it to the corresponding floating point
6541 value. If the value cannot fit in the floating point value, the results
6547 .. code-block:: llvm
6549 %X = uitofp i32 257 to float ; yields float:257.0
6550 %Y = uitofp i8 -1 to double ; yields double:255.0
6552 '``sitofp .. to``' Instruction
6553 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6560 <result> = sitofp <ty> <value> to <ty2> ; yields ty2
6565 The '``sitofp``' instruction regards ``value`` as a signed integer and
6566 converts that value to the ``ty2`` type.
6571 The '``sitofp``' instruction takes a value to cast, which must be a
6572 scalar or vector :ref:`integer <t_integer>` value, and a type to cast it to
6573 ``ty2``, which must be an :ref:`floating point <t_floating>` type. If
6574 ``ty`` is a vector integer type, ``ty2`` must be a vector floating point
6575 type with the same number of elements as ``ty``
6580 The '``sitofp``' instruction interprets its operand as a signed integer
6581 quantity and converts it to the corresponding floating point value. If
6582 the value cannot fit in the floating point value, the results are
6588 .. code-block:: llvm
6590 %X = sitofp i32 257 to float ; yields float:257.0
6591 %Y = sitofp i8 -1 to double ; yields double:-1.0
6595 '``ptrtoint .. to``' Instruction
6596 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6603 <result> = ptrtoint <ty> <value> to <ty2> ; yields ty2
6608 The '``ptrtoint``' instruction converts the pointer or a vector of
6609 pointers ``value`` to the integer (or vector of integers) type ``ty2``.
6614 The '``ptrtoint``' instruction takes a ``value`` to cast, which must be
6615 a value of type :ref:`pointer <t_pointer>` or a vector of pointers, and a
6616 type to cast it to ``ty2``, which must be an :ref:`integer <t_integer>` or
6617 a vector of integers type.
6622 The '``ptrtoint``' instruction converts ``value`` to integer type
6623 ``ty2`` by interpreting the pointer value as an integer and either
6624 truncating or zero extending that value to the size of the integer type.
6625 If ``value`` is smaller than ``ty2`` then a zero extension is done. If
6626 ``value`` is larger than ``ty2`` then a truncation is done. If they are
6627 the same size, then nothing is done (*no-op cast*) other than a type
6633 .. code-block:: llvm
6635 %X = ptrtoint i32* %P to i8 ; yields truncation on 32-bit architecture
6636 %Y = ptrtoint i32* %P to i64 ; yields zero extension on 32-bit architecture
6637 %Z = ptrtoint <4 x i32*> %P to <4 x i64>; yields vector zero extension for a vector of addresses on 32-bit architecture
6641 '``inttoptr .. to``' Instruction
6642 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6649 <result> = inttoptr <ty> <value> to <ty2> ; yields ty2
6654 The '``inttoptr``' instruction converts an integer ``value`` to a
6655 pointer type, ``ty2``.
6660 The '``inttoptr``' instruction takes an :ref:`integer <t_integer>` value to
6661 cast, and a type to cast it to, which must be a :ref:`pointer <t_pointer>`
6667 The '``inttoptr``' instruction converts ``value`` to type ``ty2`` by
6668 applying either a zero extension or a truncation depending on the size
6669 of the integer ``value``. If ``value`` is larger than the size of a
6670 pointer then a truncation is done. If ``value`` is smaller than the size
6671 of a pointer then a zero extension is done. If they are the same size,
6672 nothing is done (*no-op cast*).
6677 .. code-block:: llvm
6679 %X = inttoptr i32 255 to i32* ; yields zero extension on 64-bit architecture
6680 %Y = inttoptr i32 255 to i32* ; yields no-op on 32-bit architecture
6681 %Z = inttoptr i64 0 to i32* ; yields truncation on 32-bit architecture
6682 %Z = inttoptr <4 x i32> %G to <4 x i8*>; yields truncation of vector G to four pointers
6686 '``bitcast .. to``' Instruction
6687 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6694 <result> = bitcast <ty> <value> to <ty2> ; yields ty2
6699 The '``bitcast``' instruction converts ``value`` to type ``ty2`` without
6705 The '``bitcast``' instruction takes a value to cast, which must be a
6706 non-aggregate first class value, and a type to cast it to, which must
6707 also be a non-aggregate :ref:`first class <t_firstclass>` type. The
6708 bit sizes of ``value`` and the destination type, ``ty2``, must be
6709 identical. If the source type is a pointer, the destination type must
6710 also be a pointer of the same size. This instruction supports bitwise
6711 conversion of vectors to integers and to vectors of other types (as
6712 long as they have the same size).
6717 The '``bitcast``' instruction converts ``value`` to type ``ty2``. It
6718 is always a *no-op cast* because no bits change with this
6719 conversion. The conversion is done as if the ``value`` had been stored
6720 to memory and read back as type ``ty2``. Pointer (or vector of
6721 pointers) types may only be converted to other pointer (or vector of
6722 pointers) types with the same address space through this instruction.
6723 To convert pointers to other types, use the :ref:`inttoptr <i_inttoptr>`
6724 or :ref:`ptrtoint <i_ptrtoint>` instructions first.
6729 .. code-block:: llvm
6731 %X = bitcast i8 255 to i8 ; yields i8 :-1
6732 %Y = bitcast i32* %x to sint* ; yields sint*:%x
6733 %Z = bitcast <2 x int> %V to i64; ; yields i64: %V
6734 %Z = bitcast <2 x i32*> %V to <2 x i64*> ; yields <2 x i64*>
6736 .. _i_addrspacecast:
6738 '``addrspacecast .. to``' Instruction
6739 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6746 <result> = addrspacecast <pty> <ptrval> to <pty2> ; yields pty2
6751 The '``addrspacecast``' instruction converts ``ptrval`` from ``pty`` in
6752 address space ``n`` to type ``pty2`` in address space ``m``.
6757 The '``addrspacecast``' instruction takes a pointer or vector of pointer value
6758 to cast and a pointer type to cast it to, which must have a different
6764 The '``addrspacecast``' instruction converts the pointer value
6765 ``ptrval`` to type ``pty2``. It can be a *no-op cast* or a complex
6766 value modification, depending on the target and the address space
6767 pair. Pointer conversions within the same address space must be
6768 performed with the ``bitcast`` instruction. Note that if the address space
6769 conversion is legal then both result and operand refer to the same memory
6775 .. code-block:: llvm
6777 %X = addrspacecast i32* %x to i32 addrspace(1)* ; yields i32 addrspace(1)*:%x
6778 %Y = addrspacecast i32 addrspace(1)* %y to i64 addrspace(2)* ; yields i64 addrspace(2)*:%y
6779 %Z = addrspacecast <4 x i32*> %z to <4 x float addrspace(3)*> ; yields <4 x float addrspace(3)*>:%z
6786 The instructions in this category are the "miscellaneous" instructions,
6787 which defy better classification.
6791 '``icmp``' Instruction
6792 ^^^^^^^^^^^^^^^^^^^^^^
6799 <result> = icmp <cond> <ty> <op1>, <op2> ; yields i1 or <N x i1>:result
6804 The '``icmp``' instruction returns a boolean value or a vector of
6805 boolean values based on comparison of its two integer, integer vector,
6806 pointer, or pointer vector operands.
6811 The '``icmp``' instruction takes three operands. The first operand is
6812 the condition code indicating the kind of comparison to perform. It is
6813 not a value, just a keyword. The possible condition code are:
6816 #. ``ne``: not equal
6817 #. ``ugt``: unsigned greater than
6818 #. ``uge``: unsigned greater or equal
6819 #. ``ult``: unsigned less than
6820 #. ``ule``: unsigned less or equal
6821 #. ``sgt``: signed greater than
6822 #. ``sge``: signed greater or equal
6823 #. ``slt``: signed less than
6824 #. ``sle``: signed less or equal
6826 The remaining two arguments must be :ref:`integer <t_integer>` or
6827 :ref:`pointer <t_pointer>` or integer :ref:`vector <t_vector>` typed. They
6828 must also be identical types.
6833 The '``icmp``' compares ``op1`` and ``op2`` according to the condition
6834 code given as ``cond``. The comparison performed always yields either an
6835 :ref:`i1 <t_integer>` or vector of ``i1`` result, as follows:
6837 #. ``eq``: yields ``true`` if the operands are equal, ``false``
6838 otherwise. No sign interpretation is necessary or performed.
6839 #. ``ne``: yields ``true`` if the operands are unequal, ``false``
6840 otherwise. No sign interpretation is necessary or performed.
6841 #. ``ugt``: interprets the operands as unsigned values and yields
6842 ``true`` if ``op1`` is greater than ``op2``.
6843 #. ``uge``: interprets the operands as unsigned values and yields
6844 ``true`` if ``op1`` is greater than or equal to ``op2``.
6845 #. ``ult``: interprets the operands as unsigned values and yields
6846 ``true`` if ``op1`` is less than ``op2``.
6847 #. ``ule``: interprets the operands as unsigned values and yields
6848 ``true`` if ``op1`` is less than or equal to ``op2``.
6849 #. ``sgt``: interprets the operands as signed values and yields ``true``
6850 if ``op1`` is greater than ``op2``.
6851 #. ``sge``: interprets the operands as signed values and yields ``true``
6852 if ``op1`` is greater than or equal to ``op2``.
6853 #. ``slt``: interprets the operands as signed values and yields ``true``
6854 if ``op1`` is less than ``op2``.
6855 #. ``sle``: interprets the operands as signed values and yields ``true``
6856 if ``op1`` is less than or equal to ``op2``.
6858 If the operands are :ref:`pointer <t_pointer>` typed, the pointer values
6859 are compared as if they were integers.
6861 If the operands are integer vectors, then they are compared element by
6862 element. The result is an ``i1`` vector with the same number of elements
6863 as the values being compared. Otherwise, the result is an ``i1``.
6868 .. code-block:: llvm
6870 <result> = icmp eq i32 4, 5 ; yields: result=false
6871 <result> = icmp ne float* %X, %X ; yields: result=false
6872 <result> = icmp ult i16 4, 5 ; yields: result=true
6873 <result> = icmp sgt i16 4, 5 ; yields: result=false
6874 <result> = icmp ule i16 -4, 5 ; yields: result=false
6875 <result> = icmp sge i16 4, 5 ; yields: result=false
6877 Note that the code generator does not yet support vector types with the
6878 ``icmp`` instruction.
6882 '``fcmp``' Instruction
6883 ^^^^^^^^^^^^^^^^^^^^^^
6890 <result> = fcmp <cond> <ty> <op1>, <op2> ; yields i1 or <N x i1>:result
6895 The '``fcmp``' instruction returns a boolean value or vector of boolean
6896 values based on comparison of its operands.
6898 If the operands are floating point scalars, then the result type is a
6899 boolean (:ref:`i1 <t_integer>`).
6901 If the operands are floating point vectors, then the result type is a
6902 vector of boolean with the same number of elements as the operands being
6908 The '``fcmp``' instruction takes three operands. The first operand is
6909 the condition code indicating the kind of comparison to perform. It is
6910 not a value, just a keyword. The possible condition code are:
6912 #. ``false``: no comparison, always returns false
6913 #. ``oeq``: ordered and equal
6914 #. ``ogt``: ordered and greater than
6915 #. ``oge``: ordered and greater than or equal
6916 #. ``olt``: ordered and less than
6917 #. ``ole``: ordered and less than or equal
6918 #. ``one``: ordered and not equal
6919 #. ``ord``: ordered (no nans)
6920 #. ``ueq``: unordered or equal
6921 #. ``ugt``: unordered or greater than
6922 #. ``uge``: unordered or greater than or equal
6923 #. ``ult``: unordered or less than
6924 #. ``ule``: unordered or less than or equal
6925 #. ``une``: unordered or not equal
6926 #. ``uno``: unordered (either nans)
6927 #. ``true``: no comparison, always returns true
6929 *Ordered* means that neither operand is a QNAN while *unordered* means
6930 that either operand may be a QNAN.
6932 Each of ``val1`` and ``val2`` arguments must be either a :ref:`floating
6933 point <t_floating>` type or a :ref:`vector <t_vector>` of floating point
6934 type. They must have identical types.
6939 The '``fcmp``' instruction compares ``op1`` and ``op2`` according to the
6940 condition code given as ``cond``. If the operands are vectors, then the
6941 vectors are compared element by element. Each comparison performed
6942 always yields an :ref:`i1 <t_integer>` result, as follows:
6944 #. ``false``: always yields ``false``, regardless of operands.
6945 #. ``oeq``: yields ``true`` if both operands are not a QNAN and ``op1``
6946 is equal to ``op2``.
6947 #. ``ogt``: yields ``true`` if both operands are not a QNAN and ``op1``
6948 is greater than ``op2``.
6949 #. ``oge``: yields ``true`` if both operands are not a QNAN and ``op1``
6950 is greater than or equal to ``op2``.
6951 #. ``olt``: yields ``true`` if both operands are not a QNAN and ``op1``
6952 is less than ``op2``.
6953 #. ``ole``: yields ``true`` if both operands are not a QNAN and ``op1``
6954 is less than or equal to ``op2``.
6955 #. ``one``: yields ``true`` if both operands are not a QNAN and ``op1``
6956 is not equal to ``op2``.
6957 #. ``ord``: yields ``true`` if both operands are not a QNAN.
6958 #. ``ueq``: yields ``true`` if either operand is a QNAN or ``op1`` is
6960 #. ``ugt``: yields ``true`` if either operand is a QNAN or ``op1`` is
6961 greater than ``op2``.
6962 #. ``uge``: yields ``true`` if either operand is a QNAN or ``op1`` is
6963 greater than or equal to ``op2``.
6964 #. ``ult``: yields ``true`` if either operand is a QNAN or ``op1`` is
6966 #. ``ule``: yields ``true`` if either operand is a QNAN or ``op1`` is
6967 less than or equal to ``op2``.
6968 #. ``une``: yields ``true`` if either operand is a QNAN or ``op1`` is
6969 not equal to ``op2``.
6970 #. ``uno``: yields ``true`` if either operand is a QNAN.
6971 #. ``true``: always yields ``true``, regardless of operands.
6976 .. code-block:: llvm
6978 <result> = fcmp oeq float 4.0, 5.0 ; yields: result=false
6979 <result> = fcmp one float 4.0, 5.0 ; yields: result=true
6980 <result> = fcmp olt float 4.0, 5.0 ; yields: result=true
6981 <result> = fcmp ueq double 1.0, 2.0 ; yields: result=false
6983 Note that the code generator does not yet support vector types with the
6984 ``fcmp`` instruction.
6988 '``phi``' Instruction
6989 ^^^^^^^^^^^^^^^^^^^^^
6996 <result> = phi <ty> [ <val0>, <label0>], ...
7001 The '``phi``' instruction is used to implement the φ node in the SSA
7002 graph representing the function.
7007 The type of the incoming values is specified with the first type field.
7008 After this, the '``phi``' instruction takes a list of pairs as
7009 arguments, with one pair for each predecessor basic block of the current
7010 block. Only values of :ref:`first class <t_firstclass>` type may be used as
7011 the value arguments to the PHI node. Only labels may be used as the
7014 There must be no non-phi instructions between the start of a basic block
7015 and the PHI instructions: i.e. PHI instructions must be first in a basic
7018 For the purposes of the SSA form, the use of each incoming value is
7019 deemed to occur on the edge from the corresponding predecessor block to
7020 the current block (but after any definition of an '``invoke``'
7021 instruction's return value on the same edge).
7026 At runtime, the '``phi``' instruction logically takes on the value
7027 specified by the pair corresponding to the predecessor basic block that
7028 executed just prior to the current block.
7033 .. code-block:: llvm
7035 Loop: ; Infinite loop that counts from 0 on up...
7036 %indvar = phi i32 [ 0, %LoopHeader ], [ %nextindvar, %Loop ]
7037 %nextindvar = add i32 %indvar, 1
7042 '``select``' Instruction
7043 ^^^^^^^^^^^^^^^^^^^^^^^^
7050 <result> = select selty <cond>, <ty> <val1>, <ty> <val2> ; yields ty
7052 selty is either i1 or {<N x i1>}
7057 The '``select``' instruction is used to choose one value based on a
7058 condition, without IR-level branching.
7063 The '``select``' instruction requires an 'i1' value or a vector of 'i1'
7064 values indicating the condition, and two values of the same :ref:`first
7065 class <t_firstclass>` type.
7070 If the condition is an i1 and it evaluates to 1, the instruction returns
7071 the first value argument; otherwise, it returns the second value
7074 If the condition is a vector of i1, then the value arguments must be
7075 vectors of the same size, and the selection is done element by element.
7077 If the condition is an i1 and the value arguments are vectors of the
7078 same size, then an entire vector is selected.
7083 .. code-block:: llvm
7085 %X = select i1 true, i8 17, i8 42 ; yields i8:17
7089 '``call``' Instruction
7090 ^^^^^^^^^^^^^^^^^^^^^^
7097 <result> = [tail | musttail] call [cconv] [ret attrs] <ty> [<fnty>*] <fnptrval>(<function args>) [fn attrs]
7102 The '``call``' instruction represents a simple function call.
7107 This instruction requires several arguments:
7109 #. The optional ``tail`` and ``musttail`` markers indicate that the optimizers
7110 should perform tail call optimization. The ``tail`` marker is a hint that
7111 `can be ignored <CodeGenerator.html#sibcallopt>`_. The ``musttail`` marker
7112 means that the call must be tail call optimized in order for the program to
7113 be correct. The ``musttail`` marker provides these guarantees:
7115 #. The call will not cause unbounded stack growth if it is part of a
7116 recursive cycle in the call graph.
7117 #. Arguments with the :ref:`inalloca <attr_inalloca>` attribute are
7120 Both markers imply that the callee does not access allocas or varargs from
7121 the caller. Calls marked ``musttail`` must obey the following additional
7124 - The call must immediately precede a :ref:`ret <i_ret>` instruction,
7125 or a pointer bitcast followed by a ret instruction.
7126 - The ret instruction must return the (possibly bitcasted) value
7127 produced by the call or void.
7128 - The caller and callee prototypes must match. Pointer types of
7129 parameters or return types may differ in pointee type, but not
7131 - The calling conventions of the caller and callee must match.
7132 - All ABI-impacting function attributes, such as sret, byval, inreg,
7133 returned, and inalloca, must match.
7134 - The callee must be varargs iff the caller is varargs. Bitcasting a
7135 non-varargs function to the appropriate varargs type is legal so
7136 long as the non-varargs prefixes obey the other rules.
7138 Tail call optimization for calls marked ``tail`` is guaranteed to occur if
7139 the following conditions are met:
7141 - Caller and callee both have the calling convention ``fastcc``.
7142 - The call is in tail position (ret immediately follows call and ret
7143 uses value of call or is void).
7144 - Option ``-tailcallopt`` is enabled, or
7145 ``llvm::GuaranteedTailCallOpt`` is ``true``.
7146 - `Platform-specific constraints are
7147 met. <CodeGenerator.html#tailcallopt>`_
7149 #. The optional "cconv" marker indicates which :ref:`calling
7150 convention <callingconv>` the call should use. If none is
7151 specified, the call defaults to using C calling conventions. The
7152 calling convention of the call must match the calling convention of
7153 the target function, or else the behavior is undefined.
7154 #. The optional :ref:`Parameter Attributes <paramattrs>` list for return
7155 values. Only '``zeroext``', '``signext``', and '``inreg``' attributes
7157 #. '``ty``': the type of the call instruction itself which is also the
7158 type of the return value. Functions that return no value are marked
7160 #. '``fnty``': shall be the signature of the pointer to function value
7161 being invoked. The argument types must match the types implied by
7162 this signature. This type can be omitted if the function is not
7163 varargs and if the function type does not return a pointer to a
7165 #. '``fnptrval``': An LLVM value containing a pointer to a function to
7166 be invoked. In most cases, this is a direct function invocation, but
7167 indirect ``call``'s are just as possible, calling an arbitrary pointer
7169 #. '``function args``': argument list whose types match the function
7170 signature argument types and parameter attributes. All arguments must
7171 be of :ref:`first class <t_firstclass>` type. If the function signature
7172 indicates the function accepts a variable number of arguments, the
7173 extra arguments can be specified.
7174 #. The optional :ref:`function attributes <fnattrs>` list. Only
7175 '``noreturn``', '``nounwind``', '``readonly``' and '``readnone``'
7176 attributes are valid here.
7181 The '``call``' instruction is used to cause control flow to transfer to
7182 a specified function, with its incoming arguments bound to the specified
7183 values. Upon a '``ret``' instruction in the called function, control
7184 flow continues with the instruction after the function call, and the
7185 return value of the function is bound to the result argument.
7190 .. code-block:: llvm
7192 %retval = call i32 @test(i32 %argc)
7193 call i32 (i8*, ...)* @printf(i8* %msg, i32 12, i8 42) ; yields i32
7194 %X = tail call i32 @foo() ; yields i32
7195 %Y = tail call fastcc i32 @foo() ; yields i32
7196 call void %foo(i8 97 signext)
7198 %struct.A = type { i32, i8 }
7199 %r = call %struct.A @foo() ; yields { i32, i8 }
7200 %gr = extractvalue %struct.A %r, 0 ; yields i32
7201 %gr1 = extractvalue %struct.A %r, 1 ; yields i8
7202 %Z = call void @foo() noreturn ; indicates that %foo never returns normally
7203 %ZZ = call zeroext i32 @bar() ; Return value is %zero extended
7205 llvm treats calls to some functions with names and arguments that match
7206 the standard C99 library as being the C99 library functions, and may
7207 perform optimizations or generate code for them under that assumption.
7208 This is something we'd like to change in the future to provide better
7209 support for freestanding environments and non-C-based languages.
7213 '``va_arg``' Instruction
7214 ^^^^^^^^^^^^^^^^^^^^^^^^
7221 <resultval> = va_arg <va_list*> <arglist>, <argty>
7226 The '``va_arg``' instruction is used to access arguments passed through
7227 the "variable argument" area of a function call. It is used to implement
7228 the ``va_arg`` macro in C.
7233 This instruction takes a ``va_list*`` value and the type of the
7234 argument. It returns a value of the specified argument type and
7235 increments the ``va_list`` to point to the next argument. The actual
7236 type of ``va_list`` is target specific.
7241 The '``va_arg``' instruction loads an argument of the specified type
7242 from the specified ``va_list`` and causes the ``va_list`` to point to
7243 the next argument. For more information, see the variable argument
7244 handling :ref:`Intrinsic Functions <int_varargs>`.
7246 It is legal for this instruction to be called in a function which does
7247 not take a variable number of arguments, for example, the ``vfprintf``
7250 ``va_arg`` is an LLVM instruction instead of an :ref:`intrinsic
7251 function <intrinsics>` because it takes a type as an argument.
7256 See the :ref:`variable argument processing <int_varargs>` section.
7258 Note that the code generator does not yet fully support va\_arg on many
7259 targets. Also, it does not currently support va\_arg with aggregate
7260 types on any target.
7264 '``landingpad``' Instruction
7265 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7272 <resultval> = landingpad <resultty> personality <type> <pers_fn> <clause>+
7273 <resultval> = landingpad <resultty> personality <type> <pers_fn> cleanup <clause>*
7275 <clause> := catch <type> <value>
7276 <clause> := filter <array constant type> <array constant>
7281 The '``landingpad``' instruction is used by `LLVM's exception handling
7282 system <ExceptionHandling.html#overview>`_ to specify that a basic block
7283 is a landing pad --- one where the exception lands, and corresponds to the
7284 code found in the ``catch`` portion of a ``try``/``catch`` sequence. It
7285 defines values supplied by the personality function (``pers_fn``) upon
7286 re-entry to the function. The ``resultval`` has the type ``resultty``.
7291 This instruction takes a ``pers_fn`` value. This is the personality
7292 function associated with the unwinding mechanism. The optional
7293 ``cleanup`` flag indicates that the landing pad block is a cleanup.
7295 A ``clause`` begins with the clause type --- ``catch`` or ``filter`` --- and
7296 contains the global variable representing the "type" that may be caught
7297 or filtered respectively. Unlike the ``catch`` clause, the ``filter``
7298 clause takes an array constant as its argument. Use
7299 "``[0 x i8**] undef``" for a filter which cannot throw. The
7300 '``landingpad``' instruction must contain *at least* one ``clause`` or
7301 the ``cleanup`` flag.
7306 The '``landingpad``' instruction defines the values which are set by the
7307 personality function (``pers_fn``) upon re-entry to the function, and
7308 therefore the "result type" of the ``landingpad`` instruction. As with
7309 calling conventions, how the personality function results are
7310 represented in LLVM IR is target specific.
7312 The clauses are applied in order from top to bottom. If two
7313 ``landingpad`` instructions are merged together through inlining, the
7314 clauses from the calling function are appended to the list of clauses.
7315 When the call stack is being unwound due to an exception being thrown,
7316 the exception is compared against each ``clause`` in turn. If it doesn't
7317 match any of the clauses, and the ``cleanup`` flag is not set, then
7318 unwinding continues further up the call stack.
7320 The ``landingpad`` instruction has several restrictions:
7322 - A landing pad block is a basic block which is the unwind destination
7323 of an '``invoke``' instruction.
7324 - A landing pad block must have a '``landingpad``' instruction as its
7325 first non-PHI instruction.
7326 - There can be only one '``landingpad``' instruction within the landing
7328 - A basic block that is not a landing pad block may not include a
7329 '``landingpad``' instruction.
7330 - All '``landingpad``' instructions in a function must have the same
7331 personality function.
7336 .. code-block:: llvm
7338 ;; A landing pad which can catch an integer.
7339 %res = landingpad { i8*, i32 } personality i32 (...)* @__gxx_personality_v0
7341 ;; A landing pad that is a cleanup.
7342 %res = landingpad { i8*, i32 } personality i32 (...)* @__gxx_personality_v0
7344 ;; A landing pad which can catch an integer and can only throw a double.
7345 %res = landingpad { i8*, i32 } personality i32 (...)* @__gxx_personality_v0
7347 filter [1 x i8**] [@_ZTId]
7354 LLVM supports the notion of an "intrinsic function". These functions
7355 have well known names and semantics and are required to follow certain
7356 restrictions. Overall, these intrinsics represent an extension mechanism
7357 for the LLVM language that does not require changing all of the
7358 transformations in LLVM when adding to the language (or the bitcode
7359 reader/writer, the parser, etc...).
7361 Intrinsic function names must all start with an "``llvm.``" prefix. This
7362 prefix is reserved in LLVM for intrinsic names; thus, function names may
7363 not begin with this prefix. Intrinsic functions must always be external
7364 functions: you cannot define the body of intrinsic functions. Intrinsic
7365 functions may only be used in call or invoke instructions: it is illegal
7366 to take the address of an intrinsic function. Additionally, because
7367 intrinsic functions are part of the LLVM language, it is required if any
7368 are added that they be documented here.
7370 Some intrinsic functions can be overloaded, i.e., the intrinsic
7371 represents a family of functions that perform the same operation but on
7372 different data types. Because LLVM can represent over 8 million
7373 different integer types, overloading is used commonly to allow an
7374 intrinsic function to operate on any integer type. One or more of the
7375 argument types or the result type can be overloaded to accept any
7376 integer type. Argument types may also be defined as exactly matching a
7377 previous argument's type or the result type. This allows an intrinsic
7378 function which accepts multiple arguments, but needs all of them to be
7379 of the same type, to only be overloaded with respect to a single
7380 argument or the result.
7382 Overloaded intrinsics will have the names of its overloaded argument
7383 types encoded into its function name, each preceded by a period. Only
7384 those types which are overloaded result in a name suffix. Arguments
7385 whose type is matched against another type do not. For example, the
7386 ``llvm.ctpop`` function can take an integer of any width and returns an
7387 integer of exactly the same integer width. This leads to a family of
7388 functions such as ``i8 @llvm.ctpop.i8(i8 %val)`` and
7389 ``i29 @llvm.ctpop.i29(i29 %val)``. Only one type, the return type, is
7390 overloaded, and only one type suffix is required. Because the argument's
7391 type is matched against the return type, it does not require its own
7394 To learn how to add an intrinsic function, please see the `Extending
7395 LLVM Guide <ExtendingLLVM.html>`_.
7399 Variable Argument Handling Intrinsics
7400 -------------------------------------
7402 Variable argument support is defined in LLVM with the
7403 :ref:`va_arg <i_va_arg>` instruction and these three intrinsic
7404 functions. These functions are related to the similarly named macros
7405 defined in the ``<stdarg.h>`` header file.
7407 All of these functions operate on arguments that use a target-specific
7408 value type "``va_list``". The LLVM assembly language reference manual
7409 does not define what this type is, so all transformations should be
7410 prepared to handle these functions regardless of the type used.
7412 This example shows how the :ref:`va_arg <i_va_arg>` instruction and the
7413 variable argument handling intrinsic functions are used.
7415 .. code-block:: llvm
7417 ; This struct is different for every platform. For most platforms,
7418 ; it is merely an i8*.
7419 %struct.va_list = type { i8* }
7421 ; For Unix x86_64 platforms, va_list is the following struct:
7422 ; %struct.va_list = type { i32, i32, i8*, i8* }
7424 define i32 @test(i32 %X, ...) {
7425 ; Initialize variable argument processing
7426 %ap = alloca %struct.va_list
7427 %ap2 = bitcast %struct.va_list* %ap to i8*
7428 call void @llvm.va_start(i8* %ap2)
7430 ; Read a single integer argument
7431 %tmp = va_arg i8* %ap2, i32
7433 ; Demonstrate usage of llvm.va_copy and llvm.va_end
7435 %aq2 = bitcast i8** %aq to i8*
7436 call void @llvm.va_copy(i8* %aq2, i8* %ap2)
7437 call void @llvm.va_end(i8* %aq2)
7439 ; Stop processing of arguments.
7440 call void @llvm.va_end(i8* %ap2)
7444 declare void @llvm.va_start(i8*)
7445 declare void @llvm.va_copy(i8*, i8*)
7446 declare void @llvm.va_end(i8*)
7450 '``llvm.va_start``' Intrinsic
7451 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7458 declare void @llvm.va_start(i8* <arglist>)
7463 The '``llvm.va_start``' intrinsic initializes ``*<arglist>`` for
7464 subsequent use by ``va_arg``.
7469 The argument is a pointer to a ``va_list`` element to initialize.
7474 The '``llvm.va_start``' intrinsic works just like the ``va_start`` macro
7475 available in C. In a target-dependent way, it initializes the
7476 ``va_list`` element to which the argument points, so that the next call
7477 to ``va_arg`` will produce the first variable argument passed to the
7478 function. Unlike the C ``va_start`` macro, this intrinsic does not need
7479 to know the last argument of the function as the compiler can figure
7482 '``llvm.va_end``' Intrinsic
7483 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
7490 declare void @llvm.va_end(i8* <arglist>)
7495 The '``llvm.va_end``' intrinsic destroys ``*<arglist>``, which has been
7496 initialized previously with ``llvm.va_start`` or ``llvm.va_copy``.
7501 The argument is a pointer to a ``va_list`` to destroy.
7506 The '``llvm.va_end``' intrinsic works just like the ``va_end`` macro
7507 available in C. In a target-dependent way, it destroys the ``va_list``
7508 element to which the argument points. Calls to
7509 :ref:`llvm.va_start <int_va_start>` and
7510 :ref:`llvm.va_copy <int_va_copy>` must be matched exactly with calls to
7515 '``llvm.va_copy``' Intrinsic
7516 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7523 declare void @llvm.va_copy(i8* <destarglist>, i8* <srcarglist>)
7528 The '``llvm.va_copy``' intrinsic copies the current argument position
7529 from the source argument list to the destination argument list.
7534 The first argument is a pointer to a ``va_list`` element to initialize.
7535 The second argument is a pointer to a ``va_list`` element to copy from.
7540 The '``llvm.va_copy``' intrinsic works just like the ``va_copy`` macro
7541 available in C. In a target-dependent way, it copies the source
7542 ``va_list`` element into the destination ``va_list`` element. This
7543 intrinsic is necessary because the `` llvm.va_start`` intrinsic may be
7544 arbitrarily complex and require, for example, memory allocation.
7546 Accurate Garbage Collection Intrinsics
7547 --------------------------------------
7549 LLVM's support for `Accurate Garbage Collection <GarbageCollection.html>`_
7550 (GC) requires the frontend to generate code containing appropriate intrinsic
7551 calls and select an appropriate GC strategy which knows how to lower these
7552 intrinsics in a manner which is appropriate for the target collector.
7554 These intrinsics allow identification of :ref:`GC roots on the
7555 stack <int_gcroot>`, as well as garbage collector implementations that
7556 require :ref:`read <int_gcread>` and :ref:`write <int_gcwrite>` barriers.
7557 Frontends for type-safe garbage collected languages should generate
7558 these intrinsics to make use of the LLVM garbage collectors. For more
7559 details, see `Garbage Collection with LLVM <GarbageCollection.html>`_.
7561 Experimental Statepoint Intrinsics
7562 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7564 LLVM provides an second experimental set of intrinsics for describing garbage
7565 collection safepoints in compiled code. These intrinsics are an alternative
7566 to the ``llvm.gcroot`` intrinsics, but are compatible with the ones for
7567 :ref:`read <int_gcread>` and :ref:`write <int_gcwrite>` barriers. The
7568 differences in approach are covered in the `Garbage Collection with LLVM
7569 <GarbageCollection.html>`_ documentation. The intrinsics themselves are
7570 described in :doc:`Statepoints`.
7574 '``llvm.gcroot``' Intrinsic
7575 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
7582 declare void @llvm.gcroot(i8** %ptrloc, i8* %metadata)
7587 The '``llvm.gcroot``' intrinsic declares the existence of a GC root to
7588 the code generator, and allows some metadata to be associated with it.
7593 The first argument specifies the address of a stack object that contains
7594 the root pointer. The second pointer (which must be either a constant or
7595 a global value address) contains the meta-data to be associated with the
7601 At runtime, a call to this intrinsic stores a null pointer into the
7602 "ptrloc" location. At compile-time, the code generator generates
7603 information to allow the runtime to find the pointer at GC safe points.
7604 The '``llvm.gcroot``' intrinsic may only be used in a function which
7605 :ref:`specifies a GC algorithm <gc>`.
7609 '``llvm.gcread``' Intrinsic
7610 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
7617 declare i8* @llvm.gcread(i8* %ObjPtr, i8** %Ptr)
7622 The '``llvm.gcread``' intrinsic identifies reads of references from heap
7623 locations, allowing garbage collector implementations that require read
7629 The second argument is the address to read from, which should be an
7630 address allocated from the garbage collector. The first object is a
7631 pointer to the start of the referenced object, if needed by the language
7632 runtime (otherwise null).
7637 The '``llvm.gcread``' intrinsic has the same semantics as a load
7638 instruction, but may be replaced with substantially more complex code by
7639 the garbage collector runtime, as needed. The '``llvm.gcread``'
7640 intrinsic may only be used in a function which :ref:`specifies a GC
7645 '``llvm.gcwrite``' Intrinsic
7646 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7653 declare void @llvm.gcwrite(i8* %P1, i8* %Obj, i8** %P2)
7658 The '``llvm.gcwrite``' intrinsic identifies writes of references to heap
7659 locations, allowing garbage collector implementations that require write
7660 barriers (such as generational or reference counting collectors).
7665 The first argument is the reference to store, the second is the start of
7666 the object to store it to, and the third is the address of the field of
7667 Obj to store to. If the runtime does not require a pointer to the
7668 object, Obj may be null.
7673 The '``llvm.gcwrite``' intrinsic has the same semantics as a store
7674 instruction, but may be replaced with substantially more complex code by
7675 the garbage collector runtime, as needed. The '``llvm.gcwrite``'
7676 intrinsic may only be used in a function which :ref:`specifies a GC
7679 Code Generator Intrinsics
7680 -------------------------
7682 These intrinsics are provided by LLVM to expose special features that
7683 may only be implemented with code generator support.
7685 '``llvm.returnaddress``' Intrinsic
7686 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7693 declare i8 *@llvm.returnaddress(i32 <level>)
7698 The '``llvm.returnaddress``' intrinsic attempts to compute a
7699 target-specific value indicating the return address of the current
7700 function or one of its callers.
7705 The argument to this intrinsic indicates which function to return the
7706 address for. Zero indicates the calling function, one indicates its
7707 caller, etc. The argument is **required** to be a constant integer
7713 The '``llvm.returnaddress``' intrinsic either returns a pointer
7714 indicating the return address of the specified call frame, or zero if it
7715 cannot be identified. The value returned by this intrinsic is likely to
7716 be incorrect or 0 for arguments other than zero, so it should only be
7717 used for debugging purposes.
7719 Note that calling this intrinsic does not prevent function inlining or
7720 other aggressive transformations, so the value returned may not be that
7721 of the obvious source-language caller.
7723 '``llvm.frameaddress``' Intrinsic
7724 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7731 declare i8* @llvm.frameaddress(i32 <level>)
7736 The '``llvm.frameaddress``' intrinsic attempts to return the
7737 target-specific frame pointer value for the specified stack frame.
7742 The argument to this intrinsic indicates which function to return the
7743 frame pointer for. Zero indicates the calling function, one indicates
7744 its caller, etc. The argument is **required** to be a constant integer
7750 The '``llvm.frameaddress``' intrinsic either returns a pointer
7751 indicating the frame address of the specified call frame, or zero if it
7752 cannot be identified. The value returned by this intrinsic is likely to
7753 be incorrect or 0 for arguments other than zero, so it should only be
7754 used for debugging purposes.
7756 Note that calling this intrinsic does not prevent function inlining or
7757 other aggressive transformations, so the value returned may not be that
7758 of the obvious source-language caller.
7760 '``llvm.frameescape``' and '``llvm.framerecover``' Intrinsics
7761 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7768 declare void @llvm.frameescape(...)
7769 declare i8* @llvm.framerecover(i8* %func, i8* %fp, i32 %idx)
7774 The '``llvm.frameescape``' intrinsic escapes offsets of a collection of static
7775 allocas, and the '``llvm.framerecover``' intrinsic applies those offsets to a
7776 live frame pointer to recover the address of the allocation. The offset is
7777 computed during frame layout of the caller of ``llvm.frameescape``.
7782 All arguments to '``llvm.frameescape``' must be pointers to static allocas or
7783 casts of static allocas. Each function can only call '``llvm.frameescape``'
7784 once, and it can only do so from the entry block.
7786 The ``func`` argument to '``llvm.framerecover``' must be a constant
7787 bitcasted pointer to a function defined in the current module. The code
7788 generator cannot determine the frame allocation offset of functions defined in
7791 The ``fp`` argument to '``llvm.framerecover``' must be a frame
7792 pointer of a call frame that is currently live. The return value of
7793 '``llvm.frameaddress``' is one way to produce such a value, but most platforms
7794 also expose the frame pointer through stack unwinding mechanisms.
7796 The ``idx`` argument to '``llvm.framerecover``' indicates which alloca passed to
7797 '``llvm.frameescape``' to recover. It is zero-indexed.
7802 These intrinsics allow a group of functions to access one stack memory
7803 allocation in an ancestor stack frame. The memory returned from
7804 '``llvm.frameallocate``' may be allocated prior to stack realignment, so the
7805 memory is only aligned to the ABI-required stack alignment. Each function may
7806 only call '``llvm.frameallocate``' one or zero times from the function entry
7807 block. The frame allocation intrinsic inhibits inlining, as any frame
7808 allocations in the inlined function frame are likely to be at a different
7809 offset from the one used by '``llvm.framerecover``' called with the
7812 .. _int_read_register:
7813 .. _int_write_register:
7815 '``llvm.read_register``' and '``llvm.write_register``' Intrinsics
7816 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7823 declare i32 @llvm.read_register.i32(metadata)
7824 declare i64 @llvm.read_register.i64(metadata)
7825 declare void @llvm.write_register.i32(metadata, i32 @value)
7826 declare void @llvm.write_register.i64(metadata, i64 @value)
7832 The '``llvm.read_register``' and '``llvm.write_register``' intrinsics
7833 provides access to the named register. The register must be valid on
7834 the architecture being compiled to. The type needs to be compatible
7835 with the register being read.
7840 The '``llvm.read_register``' intrinsic returns the current value of the
7841 register, where possible. The '``llvm.write_register``' intrinsic sets
7842 the current value of the register, where possible.
7844 This is useful to implement named register global variables that need
7845 to always be mapped to a specific register, as is common practice on
7846 bare-metal programs including OS kernels.
7848 The compiler doesn't check for register availability or use of the used
7849 register in surrounding code, including inline assembly. Because of that,
7850 allocatable registers are not supported.
7852 Warning: So far it only works with the stack pointer on selected
7853 architectures (ARM, AArch64, PowerPC and x86_64). Significant amount of
7854 work is needed to support other registers and even more so, allocatable
7859 '``llvm.stacksave``' Intrinsic
7860 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7867 declare i8* @llvm.stacksave()
7872 The '``llvm.stacksave``' intrinsic is used to remember the current state
7873 of the function stack, for use with
7874 :ref:`llvm.stackrestore <int_stackrestore>`. This is useful for
7875 implementing language features like scoped automatic variable sized
7881 This intrinsic returns a opaque pointer value that can be passed to
7882 :ref:`llvm.stackrestore <int_stackrestore>`. When an
7883 ``llvm.stackrestore`` intrinsic is executed with a value saved from
7884 ``llvm.stacksave``, it effectively restores the state of the stack to
7885 the state it was in when the ``llvm.stacksave`` intrinsic executed. In
7886 practice, this pops any :ref:`alloca <i_alloca>` blocks from the stack that
7887 were allocated after the ``llvm.stacksave`` was executed.
7889 .. _int_stackrestore:
7891 '``llvm.stackrestore``' Intrinsic
7892 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7899 declare void @llvm.stackrestore(i8* %ptr)
7904 The '``llvm.stackrestore``' intrinsic is used to restore the state of
7905 the function stack to the state it was in when the corresponding
7906 :ref:`llvm.stacksave <int_stacksave>` intrinsic executed. This is
7907 useful for implementing language features like scoped automatic variable
7908 sized arrays in C99.
7913 See the description for :ref:`llvm.stacksave <int_stacksave>`.
7915 '``llvm.prefetch``' Intrinsic
7916 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7923 declare void @llvm.prefetch(i8* <address>, i32 <rw>, i32 <locality>, i32 <cache type>)
7928 The '``llvm.prefetch``' intrinsic is a hint to the code generator to
7929 insert a prefetch instruction if supported; otherwise, it is a noop.
7930 Prefetches have no effect on the behavior of the program but can change
7931 its performance characteristics.
7936 ``address`` is the address to be prefetched, ``rw`` is the specifier
7937 determining if the fetch should be for a read (0) or write (1), and
7938 ``locality`` is a temporal locality specifier ranging from (0) - no
7939 locality, to (3) - extremely local keep in cache. The ``cache type``
7940 specifies whether the prefetch is performed on the data (1) or
7941 instruction (0) cache. The ``rw``, ``locality`` and ``cache type``
7942 arguments must be constant integers.
7947 This intrinsic does not modify the behavior of the program. In
7948 particular, prefetches cannot trap and do not produce a value. On
7949 targets that support this intrinsic, the prefetch can provide hints to
7950 the processor cache for better performance.
7952 '``llvm.pcmarker``' Intrinsic
7953 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7960 declare void @llvm.pcmarker(i32 <id>)
7965 The '``llvm.pcmarker``' intrinsic is a method to export a Program
7966 Counter (PC) in a region of code to simulators and other tools. The
7967 method is target specific, but it is expected that the marker will use
7968 exported symbols to transmit the PC of the marker. The marker makes no
7969 guarantees that it will remain with any specific instruction after
7970 optimizations. It is possible that the presence of a marker will inhibit
7971 optimizations. The intended use is to be inserted after optimizations to
7972 allow correlations of simulation runs.
7977 ``id`` is a numerical id identifying the marker.
7982 This intrinsic does not modify the behavior of the program. Backends
7983 that do not support this intrinsic may ignore it.
7985 '``llvm.readcyclecounter``' Intrinsic
7986 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
7993 declare i64 @llvm.readcyclecounter()
7998 The '``llvm.readcyclecounter``' intrinsic provides access to the cycle
7999 counter register (or similar low latency, high accuracy clocks) on those
8000 targets that support it. On X86, it should map to RDTSC. On Alpha, it
8001 should map to RPCC. As the backing counters overflow quickly (on the
8002 order of 9 seconds on alpha), this should only be used for small
8008 When directly supported, reading the cycle counter should not modify any
8009 memory. Implementations are allowed to either return a application
8010 specific value or a system wide value. On backends without support, this
8011 is lowered to a constant 0.
8013 Note that runtime support may be conditional on the privilege-level code is
8014 running at and the host platform.
8016 '``llvm.clear_cache``' Intrinsic
8017 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8024 declare void @llvm.clear_cache(i8*, i8*)
8029 The '``llvm.clear_cache``' intrinsic ensures visibility of modifications
8030 in the specified range to the execution unit of the processor. On
8031 targets with non-unified instruction and data cache, the implementation
8032 flushes the instruction cache.
8037 On platforms with coherent instruction and data caches (e.g. x86), this
8038 intrinsic is a nop. On platforms with non-coherent instruction and data
8039 cache (e.g. ARM, MIPS), the intrinsic is lowered either to appropriate
8040 instructions or a system call, if cache flushing requires special
8043 The default behavior is to emit a call to ``__clear_cache`` from the run
8046 This instrinsic does *not* empty the instruction pipeline. Modifications
8047 of the current function are outside the scope of the intrinsic.
8049 '``llvm.instrprof_increment``' Intrinsic
8050 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8057 declare void @llvm.instrprof_increment(i8* <name>, i64 <hash>,
8058 i32 <num-counters>, i32 <index>)
8063 The '``llvm.instrprof_increment``' intrinsic can be emitted by a
8064 frontend for use with instrumentation based profiling. These will be
8065 lowered by the ``-instrprof`` pass to generate execution counts of a
8071 The first argument is a pointer to a global variable containing the
8072 name of the entity being instrumented. This should generally be the
8073 (mangled) function name for a set of counters.
8075 The second argument is a hash value that can be used by the consumer
8076 of the profile data to detect changes to the instrumented source, and
8077 the third is the number of counters associated with ``name``. It is an
8078 error if ``hash`` or ``num-counters`` differ between two instances of
8079 ``instrprof_increment`` that refer to the same name.
8081 The last argument refers to which of the counters for ``name`` should
8082 be incremented. It should be a value between 0 and ``num-counters``.
8087 This intrinsic represents an increment of a profiling counter. It will
8088 cause the ``-instrprof`` pass to generate the appropriate data
8089 structures and the code to increment the appropriate value, in a
8090 format that can be written out by a compiler runtime and consumed via
8091 the ``llvm-profdata`` tool.
8093 Standard C Library Intrinsics
8094 -----------------------------
8096 LLVM provides intrinsics for a few important standard C library
8097 functions. These intrinsics allow source-language front-ends to pass
8098 information about the alignment of the pointer arguments to the code
8099 generator, providing opportunity for more efficient code generation.
8103 '``llvm.memcpy``' Intrinsic
8104 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
8109 This is an overloaded intrinsic. You can use ``llvm.memcpy`` on any
8110 integer bit width and for different address spaces. Not all targets
8111 support all bit widths however.
8115 declare void @llvm.memcpy.p0i8.p0i8.i32(i8* <dest>, i8* <src>,
8116 i32 <len>, i32 <align>, i1 <isvolatile>)
8117 declare void @llvm.memcpy.p0i8.p0i8.i64(i8* <dest>, i8* <src>,
8118 i64 <len>, i32 <align>, i1 <isvolatile>)
8123 The '``llvm.memcpy.*``' intrinsics copy a block of memory from the
8124 source location to the destination location.
8126 Note that, unlike the standard libc function, the ``llvm.memcpy.*``
8127 intrinsics do not return a value, takes extra alignment/isvolatile
8128 arguments and the pointers can be in specified address spaces.
8133 The first argument is a pointer to the destination, the second is a
8134 pointer to the source. The third argument is an integer argument
8135 specifying the number of bytes to copy, the fourth argument is the
8136 alignment of the source and destination locations, and the fifth is a
8137 boolean indicating a volatile access.
8139 If the call to this intrinsic has an alignment value that is not 0 or 1,
8140 then the caller guarantees that both the source and destination pointers
8141 are aligned to that boundary.
8143 If the ``isvolatile`` parameter is ``true``, the ``llvm.memcpy`` call is
8144 a :ref:`volatile operation <volatile>`. The detailed access behavior is not
8145 very cleanly specified and it is unwise to depend on it.
8150 The '``llvm.memcpy.*``' intrinsics copy a block of memory from the
8151 source location to the destination location, which are not allowed to
8152 overlap. It copies "len" bytes of memory over. If the argument is known
8153 to be aligned to some boundary, this can be specified as the fourth
8154 argument, otherwise it should be set to 0 or 1 (both meaning no alignment).
8156 '``llvm.memmove``' Intrinsic
8157 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8162 This is an overloaded intrinsic. You can use llvm.memmove on any integer
8163 bit width and for different address space. Not all targets support all
8168 declare void @llvm.memmove.p0i8.p0i8.i32(i8* <dest>, i8* <src>,
8169 i32 <len>, i32 <align>, i1 <isvolatile>)
8170 declare void @llvm.memmove.p0i8.p0i8.i64(i8* <dest>, i8* <src>,
8171 i64 <len>, i32 <align>, i1 <isvolatile>)
8176 The '``llvm.memmove.*``' intrinsics move a block of memory from the
8177 source location to the destination location. It is similar to the
8178 '``llvm.memcpy``' intrinsic but allows the two memory locations to
8181 Note that, unlike the standard libc function, the ``llvm.memmove.*``
8182 intrinsics do not return a value, takes extra alignment/isvolatile
8183 arguments and the pointers can be in specified address spaces.
8188 The first argument is a pointer to the destination, the second is a
8189 pointer to the source. The third argument is an integer argument
8190 specifying the number of bytes to copy, the fourth argument is the
8191 alignment of the source and destination locations, and the fifth is a
8192 boolean indicating a volatile access.
8194 If the call to this intrinsic has an alignment value that is not 0 or 1,
8195 then the caller guarantees that the source and destination pointers are
8196 aligned to that boundary.
8198 If the ``isvolatile`` parameter is ``true``, the ``llvm.memmove`` call
8199 is a :ref:`volatile operation <volatile>`. The detailed access behavior is
8200 not very cleanly specified and it is unwise to depend on it.
8205 The '``llvm.memmove.*``' intrinsics copy a block of memory from the
8206 source location to the destination location, which may overlap. It
8207 copies "len" bytes of memory over. If the argument is known to be
8208 aligned to some boundary, this can be specified as the fourth argument,
8209 otherwise it should be set to 0 or 1 (both meaning no alignment).
8211 '``llvm.memset.*``' Intrinsics
8212 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8217 This is an overloaded intrinsic. You can use llvm.memset on any integer
8218 bit width and for different address spaces. However, not all targets
8219 support all bit widths.
8223 declare void @llvm.memset.p0i8.i32(i8* <dest>, i8 <val>,
8224 i32 <len>, i32 <align>, i1 <isvolatile>)
8225 declare void @llvm.memset.p0i8.i64(i8* <dest>, i8 <val>,
8226 i64 <len>, i32 <align>, i1 <isvolatile>)
8231 The '``llvm.memset.*``' intrinsics fill a block of memory with a
8232 particular byte value.
8234 Note that, unlike the standard libc function, the ``llvm.memset``
8235 intrinsic does not return a value and takes extra alignment/volatile
8236 arguments. Also, the destination can be in an arbitrary address space.
8241 The first argument is a pointer to the destination to fill, the second
8242 is the byte value with which to fill it, the third argument is an
8243 integer argument specifying the number of bytes to fill, and the fourth
8244 argument is the known alignment of the destination location.
8246 If the call to this intrinsic has an alignment value that is not 0 or 1,
8247 then the caller guarantees that the destination pointer is aligned to
8250 If the ``isvolatile`` parameter is ``true``, the ``llvm.memset`` call is
8251 a :ref:`volatile operation <volatile>`. The detailed access behavior is not
8252 very cleanly specified and it is unwise to depend on it.
8257 The '``llvm.memset.*``' intrinsics fill "len" bytes of memory starting
8258 at the destination location. If the argument is known to be aligned to
8259 some boundary, this can be specified as the fourth argument, otherwise
8260 it should be set to 0 or 1 (both meaning no alignment).
8262 '``llvm.sqrt.*``' Intrinsic
8263 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
8268 This is an overloaded intrinsic. You can use ``llvm.sqrt`` on any
8269 floating point or vector of floating point type. Not all targets support
8274 declare float @llvm.sqrt.f32(float %Val)
8275 declare double @llvm.sqrt.f64(double %Val)
8276 declare x86_fp80 @llvm.sqrt.f80(x86_fp80 %Val)
8277 declare fp128 @llvm.sqrt.f128(fp128 %Val)
8278 declare ppc_fp128 @llvm.sqrt.ppcf128(ppc_fp128 %Val)
8283 The '``llvm.sqrt``' intrinsics return the sqrt of the specified operand,
8284 returning the same value as the libm '``sqrt``' functions would. Unlike
8285 ``sqrt`` in libm, however, ``llvm.sqrt`` has undefined behavior for
8286 negative numbers other than -0.0 (which allows for better optimization,
8287 because there is no need to worry about errno being set).
8288 ``llvm.sqrt(-0.0)`` is defined to return -0.0 like IEEE sqrt.
8293 The argument and return value are floating point numbers of the same
8299 This function returns the sqrt of the specified operand if it is a
8300 nonnegative floating point number.
8302 '``llvm.powi.*``' Intrinsic
8303 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
8308 This is an overloaded intrinsic. You can use ``llvm.powi`` on any
8309 floating point or vector of floating point type. Not all targets support
8314 declare float @llvm.powi.f32(float %Val, i32 %power)
8315 declare double @llvm.powi.f64(double %Val, i32 %power)
8316 declare x86_fp80 @llvm.powi.f80(x86_fp80 %Val, i32 %power)
8317 declare fp128 @llvm.powi.f128(fp128 %Val, i32 %power)
8318 declare ppc_fp128 @llvm.powi.ppcf128(ppc_fp128 %Val, i32 %power)
8323 The '``llvm.powi.*``' intrinsics return the first operand raised to the
8324 specified (positive or negative) power. The order of evaluation of
8325 multiplications is not defined. When a vector of floating point type is
8326 used, the second argument remains a scalar integer value.
8331 The second argument is an integer power, and the first is a value to
8332 raise to that power.
8337 This function returns the first value raised to the second power with an
8338 unspecified sequence of rounding operations.
8340 '``llvm.sin.*``' Intrinsic
8341 ^^^^^^^^^^^^^^^^^^^^^^^^^^
8346 This is an overloaded intrinsic. You can use ``llvm.sin`` on any
8347 floating point or vector of floating point type. Not all targets support
8352 declare float @llvm.sin.f32(float %Val)
8353 declare double @llvm.sin.f64(double %Val)
8354 declare x86_fp80 @llvm.sin.f80(x86_fp80 %Val)
8355 declare fp128 @llvm.sin.f128(fp128 %Val)
8356 declare ppc_fp128 @llvm.sin.ppcf128(ppc_fp128 %Val)
8361 The '``llvm.sin.*``' intrinsics return the sine of the operand.
8366 The argument and return value are floating point numbers of the same
8372 This function returns the sine of the specified operand, returning the
8373 same values as the libm ``sin`` functions would, and handles error
8374 conditions in the same way.
8376 '``llvm.cos.*``' Intrinsic
8377 ^^^^^^^^^^^^^^^^^^^^^^^^^^
8382 This is an overloaded intrinsic. You can use ``llvm.cos`` on any
8383 floating point or vector of floating point type. Not all targets support
8388 declare float @llvm.cos.f32(float %Val)
8389 declare double @llvm.cos.f64(double %Val)
8390 declare x86_fp80 @llvm.cos.f80(x86_fp80 %Val)
8391 declare fp128 @llvm.cos.f128(fp128 %Val)
8392 declare ppc_fp128 @llvm.cos.ppcf128(ppc_fp128 %Val)
8397 The '``llvm.cos.*``' intrinsics return the cosine of the operand.
8402 The argument and return value are floating point numbers of the same
8408 This function returns the cosine of the specified operand, returning the
8409 same values as the libm ``cos`` functions would, and handles error
8410 conditions in the same way.
8412 '``llvm.pow.*``' Intrinsic
8413 ^^^^^^^^^^^^^^^^^^^^^^^^^^
8418 This is an overloaded intrinsic. You can use ``llvm.pow`` on any
8419 floating point or vector of floating point type. Not all targets support
8424 declare float @llvm.pow.f32(float %Val, float %Power)
8425 declare double @llvm.pow.f64(double %Val, double %Power)
8426 declare x86_fp80 @llvm.pow.f80(x86_fp80 %Val, x86_fp80 %Power)
8427 declare fp128 @llvm.pow.f128(fp128 %Val, fp128 %Power)
8428 declare ppc_fp128 @llvm.pow.ppcf128(ppc_fp128 %Val, ppc_fp128 Power)
8433 The '``llvm.pow.*``' intrinsics return the first operand raised to the
8434 specified (positive or negative) power.
8439 The second argument is a floating point power, and the first is a value
8440 to raise to that power.
8445 This function returns the first value raised to the second power,
8446 returning the same values as the libm ``pow`` functions would, and
8447 handles error conditions in the same way.
8449 '``llvm.exp.*``' Intrinsic
8450 ^^^^^^^^^^^^^^^^^^^^^^^^^^
8455 This is an overloaded intrinsic. You can use ``llvm.exp`` on any
8456 floating point or vector of floating point type. Not all targets support
8461 declare float @llvm.exp.f32(float %Val)
8462 declare double @llvm.exp.f64(double %Val)
8463 declare x86_fp80 @llvm.exp.f80(x86_fp80 %Val)
8464 declare fp128 @llvm.exp.f128(fp128 %Val)
8465 declare ppc_fp128 @llvm.exp.ppcf128(ppc_fp128 %Val)
8470 The '``llvm.exp.*``' intrinsics perform the exp function.
8475 The argument and return value are floating point numbers of the same
8481 This function returns the same values as the libm ``exp`` functions
8482 would, and handles error conditions in the same way.
8484 '``llvm.exp2.*``' Intrinsic
8485 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
8490 This is an overloaded intrinsic. You can use ``llvm.exp2`` on any
8491 floating point or vector of floating point type. Not all targets support
8496 declare float @llvm.exp2.f32(float %Val)
8497 declare double @llvm.exp2.f64(double %Val)
8498 declare x86_fp80 @llvm.exp2.f80(x86_fp80 %Val)
8499 declare fp128 @llvm.exp2.f128(fp128 %Val)
8500 declare ppc_fp128 @llvm.exp2.ppcf128(ppc_fp128 %Val)
8505 The '``llvm.exp2.*``' intrinsics perform the exp2 function.
8510 The argument and return value are floating point numbers of the same
8516 This function returns the same values as the libm ``exp2`` functions
8517 would, and handles error conditions in the same way.
8519 '``llvm.log.*``' Intrinsic
8520 ^^^^^^^^^^^^^^^^^^^^^^^^^^
8525 This is an overloaded intrinsic. You can use ``llvm.log`` on any
8526 floating point or vector of floating point type. Not all targets support
8531 declare float @llvm.log.f32(float %Val)
8532 declare double @llvm.log.f64(double %Val)
8533 declare x86_fp80 @llvm.log.f80(x86_fp80 %Val)
8534 declare fp128 @llvm.log.f128(fp128 %Val)
8535 declare ppc_fp128 @llvm.log.ppcf128(ppc_fp128 %Val)
8540 The '``llvm.log.*``' intrinsics perform the log function.
8545 The argument and return value are floating point numbers of the same
8551 This function returns the same values as the libm ``log`` functions
8552 would, and handles error conditions in the same way.
8554 '``llvm.log10.*``' Intrinsic
8555 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8560 This is an overloaded intrinsic. You can use ``llvm.log10`` on any
8561 floating point or vector of floating point type. Not all targets support
8566 declare float @llvm.log10.f32(float %Val)
8567 declare double @llvm.log10.f64(double %Val)
8568 declare x86_fp80 @llvm.log10.f80(x86_fp80 %Val)
8569 declare fp128 @llvm.log10.f128(fp128 %Val)
8570 declare ppc_fp128 @llvm.log10.ppcf128(ppc_fp128 %Val)
8575 The '``llvm.log10.*``' intrinsics perform the log10 function.
8580 The argument and return value are floating point numbers of the same
8586 This function returns the same values as the libm ``log10`` functions
8587 would, and handles error conditions in the same way.
8589 '``llvm.log2.*``' Intrinsic
8590 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
8595 This is an overloaded intrinsic. You can use ``llvm.log2`` on any
8596 floating point or vector of floating point type. Not all targets support
8601 declare float @llvm.log2.f32(float %Val)
8602 declare double @llvm.log2.f64(double %Val)
8603 declare x86_fp80 @llvm.log2.f80(x86_fp80 %Val)
8604 declare fp128 @llvm.log2.f128(fp128 %Val)
8605 declare ppc_fp128 @llvm.log2.ppcf128(ppc_fp128 %Val)
8610 The '``llvm.log2.*``' intrinsics perform the log2 function.
8615 The argument and return value are floating point numbers of the same
8621 This function returns the same values as the libm ``log2`` functions
8622 would, and handles error conditions in the same way.
8624 '``llvm.fma.*``' Intrinsic
8625 ^^^^^^^^^^^^^^^^^^^^^^^^^^
8630 This is an overloaded intrinsic. You can use ``llvm.fma`` on any
8631 floating point or vector of floating point type. Not all targets support
8636 declare float @llvm.fma.f32(float %a, float %b, float %c)
8637 declare double @llvm.fma.f64(double %a, double %b, double %c)
8638 declare x86_fp80 @llvm.fma.f80(x86_fp80 %a, x86_fp80 %b, x86_fp80 %c)
8639 declare fp128 @llvm.fma.f128(fp128 %a, fp128 %b, fp128 %c)
8640 declare ppc_fp128 @llvm.fma.ppcf128(ppc_fp128 %a, ppc_fp128 %b, ppc_fp128 %c)
8645 The '``llvm.fma.*``' intrinsics perform the fused multiply-add
8651 The argument and return value are floating point numbers of the same
8657 This function returns the same values as the libm ``fma`` functions
8658 would, and does not set errno.
8660 '``llvm.fabs.*``' Intrinsic
8661 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
8666 This is an overloaded intrinsic. You can use ``llvm.fabs`` on any
8667 floating point or vector of floating point type. Not all targets support
8672 declare float @llvm.fabs.f32(float %Val)
8673 declare double @llvm.fabs.f64(double %Val)
8674 declare x86_fp80 @llvm.fabs.f80(x86_fp80 %Val)
8675 declare fp128 @llvm.fabs.f128(fp128 %Val)
8676 declare ppc_fp128 @llvm.fabs.ppcf128(ppc_fp128 %Val)
8681 The '``llvm.fabs.*``' intrinsics return the absolute value of the
8687 The argument and return value are floating point numbers of the same
8693 This function returns the same values as the libm ``fabs`` functions
8694 would, and handles error conditions in the same way.
8696 '``llvm.minnum.*``' Intrinsic
8697 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8702 This is an overloaded intrinsic. You can use ``llvm.minnum`` on any
8703 floating point or vector of floating point type. Not all targets support
8708 declare float @llvm.minnum.f32(float %Val0, float %Val1)
8709 declare double @llvm.minnum.f64(double %Val0, double %Val1)
8710 declare x86_fp80 @llvm.minnum.f80(x86_fp80 %Val0, x86_fp80 %Val1)
8711 declare fp128 @llvm.minnum.f128(fp128 %Val0, fp128 %Val1)
8712 declare ppc_fp128 @llvm.minnum.ppcf128(ppc_fp128 %Val0, ppc_fp128 %Val1)
8717 The '``llvm.minnum.*``' intrinsics return the minimum of the two
8724 The arguments and return value are floating point numbers of the same
8730 Follows the IEEE-754 semantics for minNum, which also match for libm's
8733 If either operand is a NaN, returns the other non-NaN operand. Returns
8734 NaN only if both operands are NaN. If the operands compare equal,
8735 returns a value that compares equal to both operands. This means that
8736 fmin(+/-0.0, +/-0.0) could return either -0.0 or 0.0.
8738 '``llvm.maxnum.*``' Intrinsic
8739 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8744 This is an overloaded intrinsic. You can use ``llvm.maxnum`` on any
8745 floating point or vector of floating point type. Not all targets support
8750 declare float @llvm.maxnum.f32(float %Val0, float %Val1l)
8751 declare double @llvm.maxnum.f64(double %Val0, double %Val1)
8752 declare x86_fp80 @llvm.maxnum.f80(x86_fp80 %Val0, x86_fp80 %Val1)
8753 declare fp128 @llvm.maxnum.f128(fp128 %Val0, fp128 %Val1)
8754 declare ppc_fp128 @llvm.maxnum.ppcf128(ppc_fp128 %Val0, ppc_fp128 %Val1)
8759 The '``llvm.maxnum.*``' intrinsics return the maximum of the two
8766 The arguments and return value are floating point numbers of the same
8771 Follows the IEEE-754 semantics for maxNum, which also match for libm's
8774 If either operand is a NaN, returns the other non-NaN operand. Returns
8775 NaN only if both operands are NaN. If the operands compare equal,
8776 returns a value that compares equal to both operands. This means that
8777 fmax(+/-0.0, +/-0.0) could return either -0.0 or 0.0.
8779 '``llvm.copysign.*``' Intrinsic
8780 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8785 This is an overloaded intrinsic. You can use ``llvm.copysign`` on any
8786 floating point or vector of floating point type. Not all targets support
8791 declare float @llvm.copysign.f32(float %Mag, float %Sgn)
8792 declare double @llvm.copysign.f64(double %Mag, double %Sgn)
8793 declare x86_fp80 @llvm.copysign.f80(x86_fp80 %Mag, x86_fp80 %Sgn)
8794 declare fp128 @llvm.copysign.f128(fp128 %Mag, fp128 %Sgn)
8795 declare ppc_fp128 @llvm.copysign.ppcf128(ppc_fp128 %Mag, ppc_fp128 %Sgn)
8800 The '``llvm.copysign.*``' intrinsics return a value with the magnitude of the
8801 first operand and the sign of the second operand.
8806 The arguments and return value are floating point numbers of the same
8812 This function returns the same values as the libm ``copysign``
8813 functions would, and handles error conditions in the same way.
8815 '``llvm.floor.*``' Intrinsic
8816 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8821 This is an overloaded intrinsic. You can use ``llvm.floor`` on any
8822 floating point or vector of floating point type. Not all targets support
8827 declare float @llvm.floor.f32(float %Val)
8828 declare double @llvm.floor.f64(double %Val)
8829 declare x86_fp80 @llvm.floor.f80(x86_fp80 %Val)
8830 declare fp128 @llvm.floor.f128(fp128 %Val)
8831 declare ppc_fp128 @llvm.floor.ppcf128(ppc_fp128 %Val)
8836 The '``llvm.floor.*``' intrinsics return the floor of the operand.
8841 The argument and return value are floating point numbers of the same
8847 This function returns the same values as the libm ``floor`` functions
8848 would, and handles error conditions in the same way.
8850 '``llvm.ceil.*``' Intrinsic
8851 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
8856 This is an overloaded intrinsic. You can use ``llvm.ceil`` on any
8857 floating point or vector of floating point type. Not all targets support
8862 declare float @llvm.ceil.f32(float %Val)
8863 declare double @llvm.ceil.f64(double %Val)
8864 declare x86_fp80 @llvm.ceil.f80(x86_fp80 %Val)
8865 declare fp128 @llvm.ceil.f128(fp128 %Val)
8866 declare ppc_fp128 @llvm.ceil.ppcf128(ppc_fp128 %Val)
8871 The '``llvm.ceil.*``' intrinsics return the ceiling of the operand.
8876 The argument and return value are floating point numbers of the same
8882 This function returns the same values as the libm ``ceil`` functions
8883 would, and handles error conditions in the same way.
8885 '``llvm.trunc.*``' Intrinsic
8886 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8891 This is an overloaded intrinsic. You can use ``llvm.trunc`` on any
8892 floating point or vector of floating point type. Not all targets support
8897 declare float @llvm.trunc.f32(float %Val)
8898 declare double @llvm.trunc.f64(double %Val)
8899 declare x86_fp80 @llvm.trunc.f80(x86_fp80 %Val)
8900 declare fp128 @llvm.trunc.f128(fp128 %Val)
8901 declare ppc_fp128 @llvm.trunc.ppcf128(ppc_fp128 %Val)
8906 The '``llvm.trunc.*``' intrinsics returns the operand rounded to the
8907 nearest integer not larger in magnitude than the operand.
8912 The argument and return value are floating point numbers of the same
8918 This function returns the same values as the libm ``trunc`` functions
8919 would, and handles error conditions in the same way.
8921 '``llvm.rint.*``' Intrinsic
8922 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
8927 This is an overloaded intrinsic. You can use ``llvm.rint`` on any
8928 floating point or vector of floating point type. Not all targets support
8933 declare float @llvm.rint.f32(float %Val)
8934 declare double @llvm.rint.f64(double %Val)
8935 declare x86_fp80 @llvm.rint.f80(x86_fp80 %Val)
8936 declare fp128 @llvm.rint.f128(fp128 %Val)
8937 declare ppc_fp128 @llvm.rint.ppcf128(ppc_fp128 %Val)
8942 The '``llvm.rint.*``' intrinsics returns the operand rounded to the
8943 nearest integer. It may raise an inexact floating-point exception if the
8944 operand isn't an integer.
8949 The argument and return value are floating point numbers of the same
8955 This function returns the same values as the libm ``rint`` functions
8956 would, and handles error conditions in the same way.
8958 '``llvm.nearbyint.*``' Intrinsic
8959 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
8964 This is an overloaded intrinsic. You can use ``llvm.nearbyint`` on any
8965 floating point or vector of floating point type. Not all targets support
8970 declare float @llvm.nearbyint.f32(float %Val)
8971 declare double @llvm.nearbyint.f64(double %Val)
8972 declare x86_fp80 @llvm.nearbyint.f80(x86_fp80 %Val)
8973 declare fp128 @llvm.nearbyint.f128(fp128 %Val)
8974 declare ppc_fp128 @llvm.nearbyint.ppcf128(ppc_fp128 %Val)
8979 The '``llvm.nearbyint.*``' intrinsics returns the operand rounded to the
8985 The argument and return value are floating point numbers of the same
8991 This function returns the same values as the libm ``nearbyint``
8992 functions would, and handles error conditions in the same way.
8994 '``llvm.round.*``' Intrinsic
8995 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9000 This is an overloaded intrinsic. You can use ``llvm.round`` on any
9001 floating point or vector of floating point type. Not all targets support
9006 declare float @llvm.round.f32(float %Val)
9007 declare double @llvm.round.f64(double %Val)
9008 declare x86_fp80 @llvm.round.f80(x86_fp80 %Val)
9009 declare fp128 @llvm.round.f128(fp128 %Val)
9010 declare ppc_fp128 @llvm.round.ppcf128(ppc_fp128 %Val)
9015 The '``llvm.round.*``' intrinsics returns the operand rounded to the
9021 The argument and return value are floating point numbers of the same
9027 This function returns the same values as the libm ``round``
9028 functions would, and handles error conditions in the same way.
9030 Bit Manipulation Intrinsics
9031 ---------------------------
9033 LLVM provides intrinsics for a few important bit manipulation
9034 operations. These allow efficient code generation for some algorithms.
9036 '``llvm.bswap.*``' Intrinsics
9037 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9042 This is an overloaded intrinsic function. You can use bswap on any
9043 integer type that is an even number of bytes (i.e. BitWidth % 16 == 0).
9047 declare i16 @llvm.bswap.i16(i16 <id>)
9048 declare i32 @llvm.bswap.i32(i32 <id>)
9049 declare i64 @llvm.bswap.i64(i64 <id>)
9054 The '``llvm.bswap``' family of intrinsics is used to byte swap integer
9055 values with an even number of bytes (positive multiple of 16 bits).
9056 These are useful for performing operations on data that is not in the
9057 target's native byte order.
9062 The ``llvm.bswap.i16`` intrinsic returns an i16 value that has the high
9063 and low byte of the input i16 swapped. Similarly, the ``llvm.bswap.i32``
9064 intrinsic returns an i32 value that has the four bytes of the input i32
9065 swapped, so that if the input bytes are numbered 0, 1, 2, 3 then the
9066 returned i32 will have its bytes in 3, 2, 1, 0 order. The
9067 ``llvm.bswap.i48``, ``llvm.bswap.i64`` and other intrinsics extend this
9068 concept to additional even-byte lengths (6 bytes, 8 bytes and more,
9071 '``llvm.ctpop.*``' Intrinsic
9072 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9077 This is an overloaded intrinsic. You can use llvm.ctpop on any integer
9078 bit width, or on any vector with integer elements. Not all targets
9079 support all bit widths or vector types, however.
9083 declare i8 @llvm.ctpop.i8(i8 <src>)
9084 declare i16 @llvm.ctpop.i16(i16 <src>)
9085 declare i32 @llvm.ctpop.i32(i32 <src>)
9086 declare i64 @llvm.ctpop.i64(i64 <src>)
9087 declare i256 @llvm.ctpop.i256(i256 <src>)
9088 declare <2 x i32> @llvm.ctpop.v2i32(<2 x i32> <src>)
9093 The '``llvm.ctpop``' family of intrinsics counts the number of bits set
9099 The only argument is the value to be counted. The argument may be of any
9100 integer type, or a vector with integer elements. The return type must
9101 match the argument type.
9106 The '``llvm.ctpop``' intrinsic counts the 1's in a variable, or within
9107 each element of a vector.
9109 '``llvm.ctlz.*``' Intrinsic
9110 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
9115 This is an overloaded intrinsic. You can use ``llvm.ctlz`` on any
9116 integer bit width, or any vector whose elements are integers. Not all
9117 targets support all bit widths or vector types, however.
9121 declare i8 @llvm.ctlz.i8 (i8 <src>, i1 <is_zero_undef>)
9122 declare i16 @llvm.ctlz.i16 (i16 <src>, i1 <is_zero_undef>)
9123 declare i32 @llvm.ctlz.i32 (i32 <src>, i1 <is_zero_undef>)
9124 declare i64 @llvm.ctlz.i64 (i64 <src>, i1 <is_zero_undef>)
9125 declare i256 @llvm.ctlz.i256(i256 <src>, i1 <is_zero_undef>)
9126 declase <2 x i32> @llvm.ctlz.v2i32(<2 x i32> <src>, i1 <is_zero_undef>)
9131 The '``llvm.ctlz``' family of intrinsic functions counts the number of
9132 leading zeros in a variable.
9137 The first argument is the value to be counted. This argument may be of
9138 any integer type, or a vector with integer element type. The return
9139 type must match the first argument type.
9141 The second argument must be a constant and is a flag to indicate whether
9142 the intrinsic should ensure that a zero as the first argument produces a
9143 defined result. Historically some architectures did not provide a
9144 defined result for zero values as efficiently, and many algorithms are
9145 now predicated on avoiding zero-value inputs.
9150 The '``llvm.ctlz``' intrinsic counts the leading (most significant)
9151 zeros in a variable, or within each element of the vector. If
9152 ``src == 0`` then the result is the size in bits of the type of ``src``
9153 if ``is_zero_undef == 0`` and ``undef`` otherwise. For example,
9154 ``llvm.ctlz(i32 2) = 30``.
9156 '``llvm.cttz.*``' Intrinsic
9157 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
9162 This is an overloaded intrinsic. You can use ``llvm.cttz`` on any
9163 integer bit width, or any vector of integer elements. Not all targets
9164 support all bit widths or vector types, however.
9168 declare i8 @llvm.cttz.i8 (i8 <src>, i1 <is_zero_undef>)
9169 declare i16 @llvm.cttz.i16 (i16 <src>, i1 <is_zero_undef>)
9170 declare i32 @llvm.cttz.i32 (i32 <src>, i1 <is_zero_undef>)
9171 declare i64 @llvm.cttz.i64 (i64 <src>, i1 <is_zero_undef>)
9172 declare i256 @llvm.cttz.i256(i256 <src>, i1 <is_zero_undef>)
9173 declase <2 x i32> @llvm.cttz.v2i32(<2 x i32> <src>, i1 <is_zero_undef>)
9178 The '``llvm.cttz``' family of intrinsic functions counts the number of
9184 The first argument is the value to be counted. This argument may be of
9185 any integer type, or a vector with integer element type. The return
9186 type must match the first argument type.
9188 The second argument must be a constant and is a flag to indicate whether
9189 the intrinsic should ensure that a zero as the first argument produces a
9190 defined result. Historically some architectures did not provide a
9191 defined result for zero values as efficiently, and many algorithms are
9192 now predicated on avoiding zero-value inputs.
9197 The '``llvm.cttz``' intrinsic counts the trailing (least significant)
9198 zeros in a variable, or within each element of a vector. If ``src == 0``
9199 then the result is the size in bits of the type of ``src`` if
9200 ``is_zero_undef == 0`` and ``undef`` otherwise. For example,
9201 ``llvm.cttz(2) = 1``.
9205 Arithmetic with Overflow Intrinsics
9206 -----------------------------------
9208 LLVM provides intrinsics for some arithmetic with overflow operations.
9210 '``llvm.sadd.with.overflow.*``' Intrinsics
9211 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9216 This is an overloaded intrinsic. You can use ``llvm.sadd.with.overflow``
9217 on any integer bit width.
9221 declare {i16, i1} @llvm.sadd.with.overflow.i16(i16 %a, i16 %b)
9222 declare {i32, i1} @llvm.sadd.with.overflow.i32(i32 %a, i32 %b)
9223 declare {i64, i1} @llvm.sadd.with.overflow.i64(i64 %a, i64 %b)
9228 The '``llvm.sadd.with.overflow``' family of intrinsic functions perform
9229 a signed addition of the two arguments, and indicate whether an overflow
9230 occurred during the signed summation.
9235 The arguments (%a and %b) and the first element of the result structure
9236 may be of integer types of any bit width, but they must have the same
9237 bit width. The second element of the result structure must be of type
9238 ``i1``. ``%a`` and ``%b`` are the two values that will undergo signed
9244 The '``llvm.sadd.with.overflow``' family of intrinsic functions perform
9245 a signed addition of the two variables. They return a structure --- the
9246 first element of which is the signed summation, and the second element
9247 of which is a bit specifying if the signed summation resulted in an
9253 .. code-block:: llvm
9255 %res = call {i32, i1} @llvm.sadd.with.overflow.i32(i32 %a, i32 %b)
9256 %sum = extractvalue {i32, i1} %res, 0
9257 %obit = extractvalue {i32, i1} %res, 1
9258 br i1 %obit, label %overflow, label %normal
9260 '``llvm.uadd.with.overflow.*``' Intrinsics
9261 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9266 This is an overloaded intrinsic. You can use ``llvm.uadd.with.overflow``
9267 on any integer bit width.
9271 declare {i16, i1} @llvm.uadd.with.overflow.i16(i16 %a, i16 %b)
9272 declare {i32, i1} @llvm.uadd.with.overflow.i32(i32 %a, i32 %b)
9273 declare {i64, i1} @llvm.uadd.with.overflow.i64(i64 %a, i64 %b)
9278 The '``llvm.uadd.with.overflow``' family of intrinsic functions perform
9279 an unsigned addition of the two arguments, and indicate whether a carry
9280 occurred during the unsigned summation.
9285 The arguments (%a and %b) and the first element of the result structure
9286 may be of integer types of any bit width, but they must have the same
9287 bit width. The second element of the result structure must be of type
9288 ``i1``. ``%a`` and ``%b`` are the two values that will undergo unsigned
9294 The '``llvm.uadd.with.overflow``' family of intrinsic functions perform
9295 an unsigned addition of the two arguments. They return a structure --- the
9296 first element of which is the sum, and the second element of which is a
9297 bit specifying if the unsigned summation resulted in a carry.
9302 .. code-block:: llvm
9304 %res = call {i32, i1} @llvm.uadd.with.overflow.i32(i32 %a, i32 %b)
9305 %sum = extractvalue {i32, i1} %res, 0
9306 %obit = extractvalue {i32, i1} %res, 1
9307 br i1 %obit, label %carry, label %normal
9309 '``llvm.ssub.with.overflow.*``' Intrinsics
9310 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9315 This is an overloaded intrinsic. You can use ``llvm.ssub.with.overflow``
9316 on any integer bit width.
9320 declare {i16, i1} @llvm.ssub.with.overflow.i16(i16 %a, i16 %b)
9321 declare {i32, i1} @llvm.ssub.with.overflow.i32(i32 %a, i32 %b)
9322 declare {i64, i1} @llvm.ssub.with.overflow.i64(i64 %a, i64 %b)
9327 The '``llvm.ssub.with.overflow``' family of intrinsic functions perform
9328 a signed subtraction of the two arguments, and indicate whether an
9329 overflow occurred during the signed subtraction.
9334 The arguments (%a and %b) and the first element of the result structure
9335 may be of integer types of any bit width, but they must have the same
9336 bit width. The second element of the result structure must be of type
9337 ``i1``. ``%a`` and ``%b`` are the two values that will undergo signed
9343 The '``llvm.ssub.with.overflow``' family of intrinsic functions perform
9344 a signed subtraction of the two arguments. They return a structure --- the
9345 first element of which is the subtraction, and the second element of
9346 which is a bit specifying if the signed subtraction resulted in an
9352 .. code-block:: llvm
9354 %res = call {i32, i1} @llvm.ssub.with.overflow.i32(i32 %a, i32 %b)
9355 %sum = extractvalue {i32, i1} %res, 0
9356 %obit = extractvalue {i32, i1} %res, 1
9357 br i1 %obit, label %overflow, label %normal
9359 '``llvm.usub.with.overflow.*``' Intrinsics
9360 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9365 This is an overloaded intrinsic. You can use ``llvm.usub.with.overflow``
9366 on any integer bit width.
9370 declare {i16, i1} @llvm.usub.with.overflow.i16(i16 %a, i16 %b)
9371 declare {i32, i1} @llvm.usub.with.overflow.i32(i32 %a, i32 %b)
9372 declare {i64, i1} @llvm.usub.with.overflow.i64(i64 %a, i64 %b)
9377 The '``llvm.usub.with.overflow``' family of intrinsic functions perform
9378 an unsigned subtraction of the two arguments, and indicate whether an
9379 overflow occurred during the unsigned subtraction.
9384 The arguments (%a and %b) and the first element of the result structure
9385 may be of integer types of any bit width, but they must have the same
9386 bit width. The second element of the result structure must be of type
9387 ``i1``. ``%a`` and ``%b`` are the two values that will undergo unsigned
9393 The '``llvm.usub.with.overflow``' family of intrinsic functions perform
9394 an unsigned subtraction of the two arguments. They return a structure ---
9395 the first element of which is the subtraction, and the second element of
9396 which is a bit specifying if the unsigned subtraction resulted in an
9402 .. code-block:: llvm
9404 %res = call {i32, i1} @llvm.usub.with.overflow.i32(i32 %a, i32 %b)
9405 %sum = extractvalue {i32, i1} %res, 0
9406 %obit = extractvalue {i32, i1} %res, 1
9407 br i1 %obit, label %overflow, label %normal
9409 '``llvm.smul.with.overflow.*``' Intrinsics
9410 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9415 This is an overloaded intrinsic. You can use ``llvm.smul.with.overflow``
9416 on any integer bit width.
9420 declare {i16, i1} @llvm.smul.with.overflow.i16(i16 %a, i16 %b)
9421 declare {i32, i1} @llvm.smul.with.overflow.i32(i32 %a, i32 %b)
9422 declare {i64, i1} @llvm.smul.with.overflow.i64(i64 %a, i64 %b)
9427 The '``llvm.smul.with.overflow``' family of intrinsic functions perform
9428 a signed multiplication of the two arguments, and indicate whether an
9429 overflow occurred during the signed multiplication.
9434 The arguments (%a and %b) and the first element of the result structure
9435 may be of integer types of any bit width, but they must have the same
9436 bit width. The second element of the result structure must be of type
9437 ``i1``. ``%a`` and ``%b`` are the two values that will undergo signed
9443 The '``llvm.smul.with.overflow``' family of intrinsic functions perform
9444 a signed multiplication of the two arguments. They return a structure ---
9445 the first element of which is the multiplication, and the second element
9446 of which is a bit specifying if the signed multiplication resulted in an
9452 .. code-block:: llvm
9454 %res = call {i32, i1} @llvm.smul.with.overflow.i32(i32 %a, i32 %b)
9455 %sum = extractvalue {i32, i1} %res, 0
9456 %obit = extractvalue {i32, i1} %res, 1
9457 br i1 %obit, label %overflow, label %normal
9459 '``llvm.umul.with.overflow.*``' Intrinsics
9460 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9465 This is an overloaded intrinsic. You can use ``llvm.umul.with.overflow``
9466 on any integer bit width.
9470 declare {i16, i1} @llvm.umul.with.overflow.i16(i16 %a, i16 %b)
9471 declare {i32, i1} @llvm.umul.with.overflow.i32(i32 %a, i32 %b)
9472 declare {i64, i1} @llvm.umul.with.overflow.i64(i64 %a, i64 %b)
9477 The '``llvm.umul.with.overflow``' family of intrinsic functions perform
9478 a unsigned multiplication of the two arguments, and indicate whether an
9479 overflow occurred during the unsigned multiplication.
9484 The arguments (%a and %b) and the first element of the result structure
9485 may be of integer types of any bit width, but they must have the same
9486 bit width. The second element of the result structure must be of type
9487 ``i1``. ``%a`` and ``%b`` are the two values that will undergo unsigned
9493 The '``llvm.umul.with.overflow``' family of intrinsic functions perform
9494 an unsigned multiplication of the two arguments. They return a structure ---
9495 the first element of which is the multiplication, and the second
9496 element of which is a bit specifying if the unsigned multiplication
9497 resulted in an overflow.
9502 .. code-block:: llvm
9504 %res = call {i32, i1} @llvm.umul.with.overflow.i32(i32 %a, i32 %b)
9505 %sum = extractvalue {i32, i1} %res, 0
9506 %obit = extractvalue {i32, i1} %res, 1
9507 br i1 %obit, label %overflow, label %normal
9509 Specialised Arithmetic Intrinsics
9510 ---------------------------------
9512 '``llvm.fmuladd.*``' Intrinsic
9513 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9520 declare float @llvm.fmuladd.f32(float %a, float %b, float %c)
9521 declare double @llvm.fmuladd.f64(double %a, double %b, double %c)
9526 The '``llvm.fmuladd.*``' intrinsic functions represent multiply-add
9527 expressions that can be fused if the code generator determines that (a) the
9528 target instruction set has support for a fused operation, and (b) that the
9529 fused operation is more efficient than the equivalent, separate pair of mul
9530 and add instructions.
9535 The '``llvm.fmuladd.*``' intrinsics each take three arguments: two
9536 multiplicands, a and b, and an addend c.
9545 %0 = call float @llvm.fmuladd.f32(%a, %b, %c)
9547 is equivalent to the expression a \* b + c, except that rounding will
9548 not be performed between the multiplication and addition steps if the
9549 code generator fuses the operations. Fusion is not guaranteed, even if
9550 the target platform supports it. If a fused multiply-add is required the
9551 corresponding llvm.fma.\* intrinsic function should be used
9552 instead. This never sets errno, just as '``llvm.fma.*``'.
9557 .. code-block:: llvm
9559 %r2 = call float @llvm.fmuladd.f32(float %a, float %b, float %c) ; yields float:r2 = (a * b) + c
9561 Half Precision Floating Point Intrinsics
9562 ----------------------------------------
9564 For most target platforms, half precision floating point is a
9565 storage-only format. This means that it is a dense encoding (in memory)
9566 but does not support computation in the format.
9568 This means that code must first load the half-precision floating point
9569 value as an i16, then convert it to float with
9570 :ref:`llvm.convert.from.fp16 <int_convert_from_fp16>`. Computation can
9571 then be performed on the float value (including extending to double
9572 etc). To store the value back to memory, it is first converted to float
9573 if needed, then converted to i16 with
9574 :ref:`llvm.convert.to.fp16 <int_convert_to_fp16>`, then storing as an
9577 .. _int_convert_to_fp16:
9579 '``llvm.convert.to.fp16``' Intrinsic
9580 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9587 declare i16 @llvm.convert.to.fp16.f32(float %a)
9588 declare i16 @llvm.convert.to.fp16.f64(double %a)
9593 The '``llvm.convert.to.fp16``' intrinsic function performs a conversion from a
9594 conventional floating point type to half precision floating point format.
9599 The intrinsic function contains single argument - the value to be
9605 The '``llvm.convert.to.fp16``' intrinsic function performs a conversion from a
9606 conventional floating point format to half precision floating point format. The
9607 return value is an ``i16`` which contains the converted number.
9612 .. code-block:: llvm
9614 %res = call i16 @llvm.convert.to.fp16.f32(float %a)
9615 store i16 %res, i16* @x, align 2
9617 .. _int_convert_from_fp16:
9619 '``llvm.convert.from.fp16``' Intrinsic
9620 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9627 declare float @llvm.convert.from.fp16.f32(i16 %a)
9628 declare double @llvm.convert.from.fp16.f64(i16 %a)
9633 The '``llvm.convert.from.fp16``' intrinsic function performs a
9634 conversion from half precision floating point format to single precision
9635 floating point format.
9640 The intrinsic function contains single argument - the value to be
9646 The '``llvm.convert.from.fp16``' intrinsic function performs a
9647 conversion from half single precision floating point format to single
9648 precision floating point format. The input half-float value is
9649 represented by an ``i16`` value.
9654 .. code-block:: llvm
9656 %a = load i16, i16* @x, align 2
9657 %res = call float @llvm.convert.from.fp16(i16 %a)
9664 The LLVM debugger intrinsics (which all start with ``llvm.dbg.``
9665 prefix), are described in the `LLVM Source Level
9666 Debugging <SourceLevelDebugging.html#format_common_intrinsics>`_
9669 Exception Handling Intrinsics
9670 -----------------------------
9672 The LLVM exception handling intrinsics (which all start with
9673 ``llvm.eh.`` prefix), are described in the `LLVM Exception
9674 Handling <ExceptionHandling.html#format_common_intrinsics>`_ document.
9678 Trampoline Intrinsics
9679 ---------------------
9681 These intrinsics make it possible to excise one parameter, marked with
9682 the :ref:`nest <nest>` attribute, from a function. The result is a
9683 callable function pointer lacking the nest parameter - the caller does
9684 not need to provide a value for it. Instead, the value to use is stored
9685 in advance in a "trampoline", a block of memory usually allocated on the
9686 stack, which also contains code to splice the nest value into the
9687 argument list. This is used to implement the GCC nested function address
9690 For example, if the function is ``i32 f(i8* nest %c, i32 %x, i32 %y)``
9691 then the resulting function pointer has signature ``i32 (i32, i32)*``.
9692 It can be created as follows:
9694 .. code-block:: llvm
9696 %tramp = alloca [10 x i8], align 4 ; size and alignment only correct for X86
9697 %tramp1 = getelementptr [10 x i8], [10 x i8]* %tramp, i32 0, i32 0
9698 call i8* @llvm.init.trampoline(i8* %tramp1, i8* bitcast (i32 (i8*, i32, i32)* @f to i8*), i8* %nval)
9699 %p = call i8* @llvm.adjust.trampoline(i8* %tramp1)
9700 %fp = bitcast i8* %p to i32 (i32, i32)*
9702 The call ``%val = call i32 %fp(i32 %x, i32 %y)`` is then equivalent to
9703 ``%val = call i32 %f(i8* %nval, i32 %x, i32 %y)``.
9707 '``llvm.init.trampoline``' Intrinsic
9708 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9715 declare void @llvm.init.trampoline(i8* <tramp>, i8* <func>, i8* <nval>)
9720 This fills the memory pointed to by ``tramp`` with executable code,
9721 turning it into a trampoline.
9726 The ``llvm.init.trampoline`` intrinsic takes three arguments, all
9727 pointers. The ``tramp`` argument must point to a sufficiently large and
9728 sufficiently aligned block of memory; this memory is written to by the
9729 intrinsic. Note that the size and the alignment are target-specific -
9730 LLVM currently provides no portable way of determining them, so a
9731 front-end that generates this intrinsic needs to have some
9732 target-specific knowledge. The ``func`` argument must hold a function
9733 bitcast to an ``i8*``.
9738 The block of memory pointed to by ``tramp`` is filled with target
9739 dependent code, turning it into a function. Then ``tramp`` needs to be
9740 passed to :ref:`llvm.adjust.trampoline <int_at>` to get a pointer which can
9741 be :ref:`bitcast (to a new function) and called <int_trampoline>`. The new
9742 function's signature is the same as that of ``func`` with any arguments
9743 marked with the ``nest`` attribute removed. At most one such ``nest``
9744 argument is allowed, and it must be of pointer type. Calling the new
9745 function is equivalent to calling ``func`` with the same argument list,
9746 but with ``nval`` used for the missing ``nest`` argument. If, after
9747 calling ``llvm.init.trampoline``, the memory pointed to by ``tramp`` is
9748 modified, then the effect of any later call to the returned function
9749 pointer is undefined.
9753 '``llvm.adjust.trampoline``' Intrinsic
9754 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9761 declare i8* @llvm.adjust.trampoline(i8* <tramp>)
9766 This performs any required machine-specific adjustment to the address of
9767 a trampoline (passed as ``tramp``).
9772 ``tramp`` must point to a block of memory which already has trampoline
9773 code filled in by a previous call to
9774 :ref:`llvm.init.trampoline <int_it>`.
9779 On some architectures the address of the code to be executed needs to be
9780 different than the address where the trampoline is actually stored. This
9781 intrinsic returns the executable address corresponding to ``tramp``
9782 after performing the required machine specific adjustments. The pointer
9783 returned can then be :ref:`bitcast and executed <int_trampoline>`.
9785 .. _int_mload_mstore:
9787 Masked Vector Load and Store Intrinsics
9788 ---------------------------------------
9790 LLVM provides intrinsics for predicated vector load and store operations. The predicate is specified by a mask operand, which holds one bit per vector element, switching the associated vector lane on or off. The memory addresses corresponding to the "off" lanes are not accessed. When all bits of the mask are on, the intrinsic is identical to a regular vector load or store. When all bits are off, no memory is accessed.
9794 '``llvm.masked.load.*``' Intrinsics
9795 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9799 This is an overloaded intrinsic. The loaded data is a vector of any integer or floating point data type.
9803 declare <16 x float> @llvm.masked.load.v16f32 (<16 x float>* <ptr>, i32 <alignment>, <16 x i1> <mask>, <16 x float> <passthru>)
9804 declare <2 x double> @llvm.masked.load.v2f64 (<2 x double>* <ptr>, i32 <alignment>, <2 x i1> <mask>, <2 x double> <passthru>)
9809 Reads a vector from memory according to the provided mask. The mask holds a bit for each vector lane, and is used to prevent memory accesses to the masked-off lanes. The masked-off lanes in the result vector are taken from the corresponding lanes of the '``passthru``' operand.
9815 The first operand is the base pointer for the load. The second operand is the alignment of the source location. It must be a constant integer value. The third operand, mask, is a vector of boolean values with the same number of elements as the return type. The fourth is a pass-through value that is used to fill the masked-off lanes of the result. The return type, underlying type of the base pointer and the type of the '``passthru``' operand are the same vector types.
9821 The '``llvm.masked.load``' intrinsic is designed for conditional reading of selected vector elements in a single IR operation. It is useful for targets that support vector masked loads and allows vectorizing predicated basic blocks on these targets. Other targets may support this intrinsic differently, for example by lowering it into a sequence of branches that guard scalar load operations.
9822 The result of this operation is equivalent to a regular vector load instruction followed by a 'select' between the loaded and the passthru values, predicated on the same mask. However, using this intrinsic prevents exceptions on memory access to masked-off lanes.
9827 %res = call <16 x float> @llvm.masked.load.v16f32 (<16 x float>* %ptr, i32 4, <16 x i1>%mask, <16 x float> %passthru)
9829 ;; The result of the two following instructions is identical aside from potential memory access exception
9830 %loadlal = load <16 x float>, <16 x float>* %ptr, align 4
9831 %res = select <16 x i1> %mask, <16 x float> %loadlal, <16 x float> %passthru
9835 '``llvm.masked.store.*``' Intrinsics
9836 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9840 This is an overloaded intrinsic. The data stored in memory is a vector of any integer or floating point data type.
9844 declare void @llvm.masked.store.v8i32 (<8 x i32> <value>, <8 x i32> * <ptr>, i32 <alignment>, <8 x i1> <mask>)
9845 declare void @llvm.masked.store.v16f32(<16 x i32> <value>, <16 x i32>* <ptr>, i32 <alignment>, <16 x i1> <mask>)
9850 Writes a vector to memory according to the provided mask. The mask holds a bit for each vector lane, and is used to prevent memory accesses to the masked-off lanes.
9855 The first operand is the vector value to be written to memory. The second operand is the base pointer for the store, it has the same underlying type as the value operand. The third operand is the alignment of the destination location. The fourth operand, mask, is a vector of boolean values. The types of the mask and the value operand must have the same number of vector elements.
9861 The '``llvm.masked.store``' intrinsics is designed for conditional writing of selected vector elements in a single IR operation. It is useful for targets that support vector masked store and allows vectorizing predicated basic blocks on these targets. Other targets may support this intrinsic differently, for example by lowering it into a sequence of branches that guard scalar store operations.
9862 The result of this operation is equivalent to a load-modify-store sequence. However, using this intrinsic prevents exceptions and data races on memory access to masked-off lanes.
9866 call void @llvm.masked.store.v16f32(<16 x float> %value, <16 x float>* %ptr, i32 4, <16 x i1> %mask)
9868 ;; The result of the following instructions is identical aside from potential data races and memory access exceptions
9869 %oldval = load <16 x float>, <16 x float>* %ptr, align 4
9870 %res = select <16 x i1> %mask, <16 x float> %value, <16 x float> %oldval
9871 store <16 x float> %res, <16 x float>* %ptr, align 4
9874 Masked Vector Gather and Scatter Intrinsics
9875 -------------------------------------------
9877 LLVM provides intrinsics for vector gather and scatter operations. They are similar to :ref:`Masked Vector Load and Store <int_mload_mstore>`, except they are designed for arbitrary memory accesses, rather than sequential memory accesses. Gather and scatter also employ a mask operand, which holds one bit per vector element, switching the associated vector lane on or off. The memory addresses corresponding to the "off" lanes are not accessed. When all bits are off, no memory is accessed.
9881 '``llvm.masked.gather.*``' Intrinsics
9882 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9886 This is an overloaded intrinsic. The loaded data are multiple scalar values of any integer or floating point data type gathered together into one vector.
9890 declare <16 x float> @llvm.masked.gather.v16f32 (<16 x float*> <ptrs>, i32 <alignment>, <16 x i1> <mask>, <16 x float> <passthru>)
9891 declare <2 x double> @llvm.masked.gather.v2f64 (<2 x double*> <ptrs>, i32 <alignment>, <2 x i1> <mask>, <2 x double> <passthru>)
9896 Reads scalar values from arbitrary memory locations and gathers them into one vector. The memory locations are provided in the vector of pointers '``ptrs``'. The memory is accessed according to the provided mask. The mask holds a bit for each vector lane, and is used to prevent memory accesses to the masked-off lanes. The masked-off lanes in the result vector are taken from the corresponding lanes of the '``passthru``' operand.
9902 The first operand is a vector of pointers which holds all memory addresses to read. The second operand is an alignment of the source addresses. It must be a constant integer value. The third operand, mask, is a vector of boolean values with the same number of elements as the return type. The fourth is a pass-through value that is used to fill the masked-off lanes of the result. The return type, underlying type of the vector of pointers and the type of the '``passthru``' operand are the same vector types.
9908 The '``llvm.masked.gather``' intrinsic is designed for conditional reading of multiple scalar values from arbitrary memory locations in a single IR operation. It is useful for targets that support vector masked gathers and allows vectorizing basic blocks with data and control divergence. Other targets may support this intrinsic differently, for example by lowering it into a sequence of scalar load operations.
9909 The semantics of this operation are equivalent to a sequence of conditional scalar loads with subsequent gathering all loaded values into a single vector. The mask restricts memory access to certain lanes and facilitates vectorization of predicated basic blocks.
9914 %res = call <4 x double> @llvm.masked.gather.v4f64 (<4 x double*> %ptrs, i32 8, <4 x i1>%mask, <4 x double> <true, true, true, true>)
9916 ;; The gather with all-true mask is equivalent to the following instruction sequence
9917 %ptr0 = extractelement <4 x double*> %ptrs, i32 0
9918 %ptr1 = extractelement <4 x double*> %ptrs, i32 1
9919 %ptr2 = extractelement <4 x double*> %ptrs, i32 2
9920 %ptr3 = extractelement <4 x double*> %ptrs, i32 3
9922 %val0 = load double, double* %ptr0, align 8
9923 %val1 = load double, double* %ptr1, align 8
9924 %val2 = load double, double* %ptr2, align 8
9925 %val3 = load double, double* %ptr3, align 8
9927 %vec0 = insertelement <4 x double>undef, %val0, 0
9928 %vec01 = insertelement <4 x double>%vec0, %val1, 1
9929 %vec012 = insertelement <4 x double>%vec01, %val2, 2
9930 %vec0123 = insertelement <4 x double>%vec012, %val3, 3
9934 '``llvm.masked.scatter.*``' Intrinsics
9935 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9939 This is an overloaded intrinsic. The data stored in memory is a vector of any integer or floating point data type. Each vector element is stored in an arbitrary memory addresses. Scatter with overlapping addresses is guaranteed to be ordered from least-significant to most-significant element.
9943 declare void @llvm.masked.scatter.v8i32 (<8 x i32> <value>, <8 x i32*> <ptrs>, i32 <alignment>, <8 x i1> <mask>)
9944 declare void @llvm.masked.scatter.v16f32(<16 x i32> <value>, <16 x i32*> <ptrs>, i32 <alignment>, <16 x i1> <mask>)
9949 Writes each element from the value vector to the corresponding memory address. The memory addresses are represented as a vector of pointers. Writing is done according to the provided mask. The mask holds a bit for each vector lane, and is used to prevent memory accesses to the masked-off lanes.
9954 The first operand is a vector value to be written to memory. The second operand is a vector of pointers, pointing to where the value elements should be stored. It has the same underlying type as the value operand. The third operand is an alignment of the destination addresses. The fourth operand, mask, is a vector of boolean values. The types of the mask and the value operand must have the same number of vector elements.
9960 The '``llvm.masked.scatter``' intrinsics is designed for writing selected vector elements to arbitrary memory addresses in a single IR operation. The operation may be conditional, when not all bits in the mask are switched on. It is useful for targets that support vector masked scatter and allows vectorizing basic blocks with data and control divergency. Other targets may support this intrinsic differently, for example by lowering it into a sequence of branches that guard scalar store operations.
9964 ;; This instruction unconditionaly stores data vector in multiple addresses
9965 call @llvm.masked.scatter.v8i32 (<8 x i32> %value, <8 x i32*> %ptrs, i32 4, <8 x i1> <true, true, .. true>)
9967 ;; It is equivalent to a list of scalar stores
9968 %val0 = extractelement <8 x i32> %value, i32 0
9969 %val1 = extractelement <8 x i32> %value, i32 1
9971 %val7 = extractelement <8 x i32> %value, i32 7
9972 %ptr0 = extractelement <8 x i32*> %ptrs, i32 0
9973 %ptr1 = extractelement <8 x i32*> %ptrs, i32 1
9975 %ptr7 = extractelement <8 x i32*> %ptrs, i32 7
9976 ;; Note: the order of the following stores is important when they overlap:
9977 store i32 %val0, i32* %ptr0, align 4
9978 store i32 %val1, i32* %ptr1, align 4
9980 store i32 %val7, i32* %ptr7, align 4
9986 This class of intrinsics provides information about the lifetime of
9987 memory objects and ranges where variables are immutable.
9991 '``llvm.lifetime.start``' Intrinsic
9992 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
9999 declare void @llvm.lifetime.start(i64 <size>, i8* nocapture <ptr>)
10004 The '``llvm.lifetime.start``' intrinsic specifies the start of a memory
10010 The first argument is a constant integer representing the size of the
10011 object, or -1 if it is variable sized. The second argument is a pointer
10017 This intrinsic indicates that before this point in the code, the value
10018 of the memory pointed to by ``ptr`` is dead. This means that it is known
10019 to never be used and has an undefined value. A load from the pointer
10020 that precedes this intrinsic can be replaced with ``'undef'``.
10024 '``llvm.lifetime.end``' Intrinsic
10025 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
10032 declare void @llvm.lifetime.end(i64 <size>, i8* nocapture <ptr>)
10037 The '``llvm.lifetime.end``' intrinsic specifies the end of a memory
10043 The first argument is a constant integer representing the size of the
10044 object, or -1 if it is variable sized. The second argument is a pointer
10050 This intrinsic indicates that after this point in the code, the value of
10051 the memory pointed to by ``ptr`` is dead. This means that it is known to
10052 never be used and has an undefined value. Any stores into the memory
10053 object following this intrinsic may be removed as dead.
10055 '``llvm.invariant.start``' Intrinsic
10056 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
10063 declare {}* @llvm.invariant.start(i64 <size>, i8* nocapture <ptr>)
10068 The '``llvm.invariant.start``' intrinsic specifies that the contents of
10069 a memory object will not change.
10074 The first argument is a constant integer representing the size of the
10075 object, or -1 if it is variable sized. The second argument is a pointer
10081 This intrinsic indicates that until an ``llvm.invariant.end`` that uses
10082 the return value, the referenced memory location is constant and
10085 '``llvm.invariant.end``' Intrinsic
10086 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
10093 declare void @llvm.invariant.end({}* <start>, i64 <size>, i8* nocapture <ptr>)
10098 The '``llvm.invariant.end``' intrinsic specifies that the contents of a
10099 memory object are mutable.
10104 The first argument is the matching ``llvm.invariant.start`` intrinsic.
10105 The second argument is a constant integer representing the size of the
10106 object, or -1 if it is variable sized and the third argument is a
10107 pointer to the object.
10112 This intrinsic indicates that the memory is mutable again.
10117 This class of intrinsics is designed to be generic and has no specific
10120 '``llvm.var.annotation``' Intrinsic
10121 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
10128 declare void @llvm.var.annotation(i8* <val>, i8* <str>, i8* <str>, i32 <int>)
10133 The '``llvm.var.annotation``' intrinsic.
10138 The first argument is a pointer to a value, the second is a pointer to a
10139 global string, the third is a pointer to a global string which is the
10140 source file name, and the last argument is the line number.
10145 This intrinsic allows annotation of local variables with arbitrary
10146 strings. This can be useful for special purpose optimizations that want
10147 to look for these annotations. These have no other defined use; they are
10148 ignored by code generation and optimization.
10150 '``llvm.ptr.annotation.*``' Intrinsic
10151 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
10156 This is an overloaded intrinsic. You can use '``llvm.ptr.annotation``' on a
10157 pointer to an integer of any width. *NOTE* you must specify an address space for
10158 the pointer. The identifier for the default address space is the integer
10163 declare i8* @llvm.ptr.annotation.p<address space>i8(i8* <val>, i8* <str>, i8* <str>, i32 <int>)
10164 declare i16* @llvm.ptr.annotation.p<address space>i16(i16* <val>, i8* <str>, i8* <str>, i32 <int>)
10165 declare i32* @llvm.ptr.annotation.p<address space>i32(i32* <val>, i8* <str>, i8* <str>, i32 <int>)
10166 declare i64* @llvm.ptr.annotation.p<address space>i64(i64* <val>, i8* <str>, i8* <str>, i32 <int>)
10167 declare i256* @llvm.ptr.annotation.p<address space>i256(i256* <val>, i8* <str>, i8* <str>, i32 <int>)
10172 The '``llvm.ptr.annotation``' intrinsic.
10177 The first argument is a pointer to an integer value of arbitrary bitwidth
10178 (result of some expression), the second is a pointer to a global string, the
10179 third is a pointer to a global string which is the source file name, and the
10180 last argument is the line number. It returns the value of the first argument.
10185 This intrinsic allows annotation of a pointer to an integer with arbitrary
10186 strings. This can be useful for special purpose optimizations that want to look
10187 for these annotations. These have no other defined use; they are ignored by code
10188 generation and optimization.
10190 '``llvm.annotation.*``' Intrinsic
10191 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
10196 This is an overloaded intrinsic. You can use '``llvm.annotation``' on
10197 any integer bit width.
10201 declare i8 @llvm.annotation.i8(i8 <val>, i8* <str>, i8* <str>, i32 <int>)
10202 declare i16 @llvm.annotation.i16(i16 <val>, i8* <str>, i8* <str>, i32 <int>)
10203 declare i32 @llvm.annotation.i32(i32 <val>, i8* <str>, i8* <str>, i32 <int>)
10204 declare i64 @llvm.annotation.i64(i64 <val>, i8* <str>, i8* <str>, i32 <int>)
10205 declare i256 @llvm.annotation.i256(i256 <val>, i8* <str>, i8* <str>, i32 <int>)
10210 The '``llvm.annotation``' intrinsic.
10215 The first argument is an integer value (result of some expression), the
10216 second is a pointer to a global string, the third is a pointer to a
10217 global string which is the source file name, and the last argument is
10218 the line number. It returns the value of the first argument.
10223 This intrinsic allows annotations to be put on arbitrary expressions
10224 with arbitrary strings. This can be useful for special purpose
10225 optimizations that want to look for these annotations. These have no
10226 other defined use; they are ignored by code generation and optimization.
10228 '``llvm.trap``' Intrinsic
10229 ^^^^^^^^^^^^^^^^^^^^^^^^^
10236 declare void @llvm.trap() noreturn nounwind
10241 The '``llvm.trap``' intrinsic.
10251 This intrinsic is lowered to the target dependent trap instruction. If
10252 the target does not have a trap instruction, this intrinsic will be
10253 lowered to a call of the ``abort()`` function.
10255 '``llvm.debugtrap``' Intrinsic
10256 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
10263 declare void @llvm.debugtrap() nounwind
10268 The '``llvm.debugtrap``' intrinsic.
10278 This intrinsic is lowered to code which is intended to cause an
10279 execution trap with the intention of requesting the attention of a
10282 '``llvm.stackprotector``' Intrinsic
10283 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
10290 declare void @llvm.stackprotector(i8* <guard>, i8** <slot>)
10295 The ``llvm.stackprotector`` intrinsic takes the ``guard`` and stores it
10296 onto the stack at ``slot``. The stack slot is adjusted to ensure that it
10297 is placed on the stack before local variables.
10302 The ``llvm.stackprotector`` intrinsic requires two pointer arguments.
10303 The first argument is the value loaded from the stack guard
10304 ``@__stack_chk_guard``. The second variable is an ``alloca`` that has
10305 enough space to hold the value of the guard.
10310 This intrinsic causes the prologue/epilogue inserter to force the position of
10311 the ``AllocaInst`` stack slot to be before local variables on the stack. This is
10312 to ensure that if a local variable on the stack is overwritten, it will destroy
10313 the value of the guard. When the function exits, the guard on the stack is
10314 checked against the original guard by ``llvm.stackprotectorcheck``. If they are
10315 different, then ``llvm.stackprotectorcheck`` causes the program to abort by
10316 calling the ``__stack_chk_fail()`` function.
10318 '``llvm.stackprotectorcheck``' Intrinsic
10319 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
10326 declare void @llvm.stackprotectorcheck(i8** <guard>)
10331 The ``llvm.stackprotectorcheck`` intrinsic compares ``guard`` against an already
10332 created stack protector and if they are not equal calls the
10333 ``__stack_chk_fail()`` function.
10338 The ``llvm.stackprotectorcheck`` intrinsic requires one pointer argument, the
10339 the variable ``@__stack_chk_guard``.
10344 This intrinsic is provided to perform the stack protector check by comparing
10345 ``guard`` with the stack slot created by ``llvm.stackprotector`` and if the
10346 values do not match call the ``__stack_chk_fail()`` function.
10348 The reason to provide this as an IR level intrinsic instead of implementing it
10349 via other IR operations is that in order to perform this operation at the IR
10350 level without an intrinsic, one would need to create additional basic blocks to
10351 handle the success/failure cases. This makes it difficult to stop the stack
10352 protector check from disrupting sibling tail calls in Codegen. With this
10353 intrinsic, we are able to generate the stack protector basic blocks late in
10354 codegen after the tail call decision has occurred.
10356 '``llvm.objectsize``' Intrinsic
10357 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
10364 declare i32 @llvm.objectsize.i32(i8* <object>, i1 <min>)
10365 declare i64 @llvm.objectsize.i64(i8* <object>, i1 <min>)
10370 The ``llvm.objectsize`` intrinsic is designed to provide information to
10371 the optimizers to determine at compile time whether a) an operation
10372 (like memcpy) will overflow a buffer that corresponds to an object, or
10373 b) that a runtime check for overflow isn't necessary. An object in this
10374 context means an allocation of a specific class, structure, array, or
10380 The ``llvm.objectsize`` intrinsic takes two arguments. The first
10381 argument is a pointer to or into the ``object``. The second argument is
10382 a boolean and determines whether ``llvm.objectsize`` returns 0 (if true)
10383 or -1 (if false) when the object size is unknown. The second argument
10384 only accepts constants.
10389 The ``llvm.objectsize`` intrinsic is lowered to a constant representing
10390 the size of the object concerned. If the size cannot be determined at
10391 compile time, ``llvm.objectsize`` returns ``i32/i64 -1 or 0`` (depending
10392 on the ``min`` argument).
10394 '``llvm.expect``' Intrinsic
10395 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
10400 This is an overloaded intrinsic. You can use ``llvm.expect`` on any
10405 declare i1 @llvm.expect.i1(i1 <val>, i1 <expected_val>)
10406 declare i32 @llvm.expect.i32(i32 <val>, i32 <expected_val>)
10407 declare i64 @llvm.expect.i64(i64 <val>, i64 <expected_val>)
10412 The ``llvm.expect`` intrinsic provides information about expected (the
10413 most probable) value of ``val``, which can be used by optimizers.
10418 The ``llvm.expect`` intrinsic takes two arguments. The first argument is
10419 a value. The second argument is an expected value, this needs to be a
10420 constant value, variables are not allowed.
10425 This intrinsic is lowered to the ``val``.
10429 '``llvm.assume``' Intrinsic
10430 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
10437 declare void @llvm.assume(i1 %cond)
10442 The ``llvm.assume`` allows the optimizer to assume that the provided
10443 condition is true. This information can then be used in simplifying other parts
10449 The condition which the optimizer may assume is always true.
10454 The intrinsic allows the optimizer to assume that the provided condition is
10455 always true whenever the control flow reaches the intrinsic call. No code is
10456 generated for this intrinsic, and instructions that contribute only to the
10457 provided condition are not used for code generation. If the condition is
10458 violated during execution, the behavior is undefined.
10460 Note that the optimizer might limit the transformations performed on values
10461 used by the ``llvm.assume`` intrinsic in order to preserve the instructions
10462 only used to form the intrinsic's input argument. This might prove undesirable
10463 if the extra information provided by the ``llvm.assume`` intrinsic does not cause
10464 sufficient overall improvement in code quality. For this reason,
10465 ``llvm.assume`` should not be used to document basic mathematical invariants
10466 that the optimizer can otherwise deduce or facts that are of little use to the
10471 '``llvm.bitset.test``' Intrinsic
10472 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
10479 declare i1 @llvm.bitset.test(i8* %ptr, metadata %bitset) nounwind readnone
10485 The first argument is a pointer to be tested. The second argument is a
10486 metadata string containing the name of a :doc:`bitset <BitSets>`.
10491 The ``llvm.bitset.test`` intrinsic tests whether the given pointer is a
10492 member of the given bitset.
10494 '``llvm.donothing``' Intrinsic
10495 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
10502 declare void @llvm.donothing() nounwind readnone
10507 The ``llvm.donothing`` intrinsic doesn't perform any operation. It's one of only
10508 two intrinsics (besides ``llvm.experimental.patchpoint``) that can be called
10509 with an invoke instruction.
10519 This intrinsic does nothing, and it's removed by optimizers and ignored
10522 Stack Map Intrinsics
10523 --------------------
10525 LLVM provides experimental intrinsics to support runtime patching
10526 mechanisms commonly desired in dynamic language JITs. These intrinsics
10527 are described in :doc:`StackMaps`.